summaryrefslogtreecommitdiff
path: root/pkg/specgen/generate
diff options
context:
space:
mode:
Diffstat (limited to 'pkg/specgen/generate')
-rw-r--r--pkg/specgen/generate/config_linux.go14
-rw-r--r--pkg/specgen/generate/container_create.go31
-rw-r--r--pkg/specgen/generate/kube/kube.go326
-rw-r--r--pkg/specgen/generate/kube/play_test.go172
-rw-r--r--pkg/specgen/generate/kube/seccomp.go84
-rw-r--r--pkg/specgen/generate/kube/volume.go124
-rw-r--r--pkg/specgen/generate/namespaces.go13
-rw-r--r--pkg/specgen/generate/security.go10
-rw-r--r--pkg/specgen/generate/storage.go61
-rw-r--r--pkg/specgen/generate/validate.go65
10 files changed, 858 insertions, 42 deletions
diff --git a/pkg/specgen/generate/config_linux.go b/pkg/specgen/generate/config_linux.go
index fac02ad01..2d40dba8f 100644
--- a/pkg/specgen/generate/config_linux.go
+++ b/pkg/specgen/generate/config_linux.go
@@ -52,7 +52,7 @@ func addPrivilegedDevices(g *generate.Generator) error {
if err == unix.EPERM {
continue
}
- return errors.Wrapf(err, "stat %s", d.Path)
+ return err
}
// Skip devices that the user has not access to.
if st.Mode()&0007 == 0 {
@@ -90,7 +90,7 @@ func DevicesFromPath(g *generate.Generator, devicePath string) error {
}
st, err := os.Stat(resolvedDevicePath)
if err != nil {
- return errors.Wrapf(err, "cannot stat device path %s", devicePath)
+ return err
}
if st.IsDir() {
found := false
@@ -231,10 +231,7 @@ func addDevice(g *generate.Generator, device string) error {
}
if rootless.IsRootless() {
if _, err := os.Stat(src); err != nil {
- if os.IsNotExist(err) {
- return errors.Wrapf(err, "the specified device %s doesn't exist", src)
- }
- return errors.Wrapf(err, "stat device %s exist", src)
+ return err
}
perm := "ro"
if strings.Contains(permissions, "w") {
@@ -353,3 +350,8 @@ func deviceFromPath(path string) (*spec.LinuxDevice, error) {
Minor: int64(unix.Minor(devNumber)),
}, nil
}
+
+func supportAmbientCapabilities() bool {
+ err := unix.Prctl(unix.PR_CAP_AMBIENT, unix.PR_CAP_AMBIENT_IS_SET, 0, 0, 0)
+ return err == nil
+}
diff --git a/pkg/specgen/generate/container_create.go b/pkg/specgen/generate/container_create.go
index f051537de..45a374216 100644
--- a/pkg/specgen/generate/container_create.go
+++ b/pkg/specgen/generate/container_create.go
@@ -111,7 +111,7 @@ func MakeContainer(ctx context.Context, rt *libpod.Runtime, s *specgen.SpecGener
return nil, errors.Wrap(err, "invalid config provided")
}
- finalMounts, finalVolumes, err := finalizeMounts(ctx, s, rt, rtc, newImage)
+ finalMounts, finalVolumes, finalOverlays, err := finalizeMounts(ctx, s, rt, rtc, newImage)
if err != nil {
return nil, err
}
@@ -121,7 +121,7 @@ func MakeContainer(ctx context.Context, rt *libpod.Runtime, s *specgen.SpecGener
return nil, err
}
- opts, err := createContainerOptions(ctx, rt, s, pod, finalVolumes, newImage, command)
+ opts, err := createContainerOptions(ctx, rt, s, pod, finalVolumes, finalOverlays, newImage, command)
if err != nil {
return nil, err
}
@@ -133,6 +133,10 @@ func MakeContainer(ctx context.Context, rt *libpod.Runtime, s *specgen.SpecGener
}
options = append(options, libpod.WithExitCommand(exitCommandArgs))
+ if len(s.Aliases) > 0 {
+ options = append(options, libpod.WithNetworkAliases(s.Aliases))
+ }
+
runtimeSpec, err := SpecGenToOCI(ctx, s, rt, rtc, newImage, finalMounts, pod, command)
if err != nil {
return nil, err
@@ -140,7 +144,7 @@ func MakeContainer(ctx context.Context, rt *libpod.Runtime, s *specgen.SpecGener
return rt.NewContainer(ctx, runtimeSpec, options...)
}
-func createContainerOptions(ctx context.Context, rt *libpod.Runtime, s *specgen.SpecGenerator, pod *libpod.Pod, volumes []*specgen.NamedVolume, img *image.Image, command []string) ([]libpod.CtrCreateOption, error) {
+func createContainerOptions(ctx context.Context, rt *libpod.Runtime, s *specgen.SpecGenerator, pod *libpod.Pod, volumes []*specgen.NamedVolume, overlays []*specgen.OverlayVolume, img *image.Image, command []string) ([]libpod.CtrCreateOption, error) {
var options []libpod.CtrCreateOption
var err error
@@ -220,9 +224,12 @@ func createContainerOptions(ctx context.Context, rt *libpod.Runtime, s *specgen.
for _, volume := range volumes {
destinations = append(destinations, volume.Dest)
}
- for _, overlayVolume := range s.OverlayVolumes {
+ for _, overlayVolume := range overlays {
destinations = append(destinations, overlayVolume.Destination)
}
+ for _, imageVolume := range s.ImageVolumes {
+ destinations = append(destinations, imageVolume.Destination)
+ }
options = append(options, libpod.WithUserVolumes(destinations))
if len(volumes) != 0 {
@@ -237,9 +244,9 @@ func createContainerOptions(ctx context.Context, rt *libpod.Runtime, s *specgen.
options = append(options, libpod.WithNamedVolumes(vols))
}
- if len(s.OverlayVolumes) != 0 {
+ if len(overlays) != 0 {
var vols []*libpod.ContainerOverlayVolume
- for _, v := range s.OverlayVolumes {
+ for _, v := range overlays {
vols = append(vols, &libpod.ContainerOverlayVolume{
Dest: v.Destination,
Source: v.Source,
@@ -248,6 +255,18 @@ func createContainerOptions(ctx context.Context, rt *libpod.Runtime, s *specgen.
options = append(options, libpod.WithOverlayVolumes(vols))
}
+ if len(s.ImageVolumes) != 0 {
+ var vols []*libpod.ContainerImageVolume
+ for _, v := range s.ImageVolumes {
+ vols = append(vols, &libpod.ContainerImageVolume{
+ Dest: v.Destination,
+ Source: v.Source,
+ ReadWrite: v.ReadWrite,
+ })
+ }
+ options = append(options, libpod.WithImageVolumes(vols))
+ }
+
if s.Command != nil {
options = append(options, libpod.WithCommand(s.Command))
}
diff --git a/pkg/specgen/generate/kube/kube.go b/pkg/specgen/generate/kube/kube.go
new file mode 100644
index 000000000..5f72d28bb
--- /dev/null
+++ b/pkg/specgen/generate/kube/kube.go
@@ -0,0 +1,326 @@
+package kube
+
+import (
+ "context"
+ "fmt"
+ "strings"
+
+ "github.com/containers/buildah/pkg/parse"
+ "github.com/containers/podman/v2/libpod/image"
+ ann "github.com/containers/podman/v2/pkg/annotations"
+ "github.com/containers/podman/v2/pkg/specgen"
+ "github.com/containers/podman/v2/pkg/util"
+ spec "github.com/opencontainers/runtime-spec/specs-go"
+ "github.com/pkg/errors"
+ v1 "k8s.io/api/core/v1"
+ "k8s.io/apimachinery/pkg/api/resource"
+)
+
+func ToPodGen(ctx context.Context, podName string, podYAML *v1.PodTemplateSpec) (*specgen.PodSpecGenerator, error) {
+ p := specgen.NewPodSpecGenerator()
+ p.Name = podName
+ p.Labels = podYAML.ObjectMeta.Labels
+ // TODO we only configure Process namespace. We also need to account for Host{IPC,Network,PID}
+ // which is not currently possible with pod create
+ if podYAML.Spec.ShareProcessNamespace != nil && *podYAML.Spec.ShareProcessNamespace {
+ p.SharedNamespaces = append(p.SharedNamespaces, "pid")
+ }
+ p.Hostname = podYAML.Spec.Hostname
+ if p.Hostname == "" {
+ p.Hostname = podName
+ }
+ if podYAML.Spec.HostNetwork {
+ p.NetNS.Value = "host"
+ }
+ if podYAML.Spec.HostAliases != nil {
+ hosts := make([]string, 0, len(podYAML.Spec.HostAliases))
+ for _, hostAlias := range podYAML.Spec.HostAliases {
+ for _, host := range hostAlias.Hostnames {
+ hosts = append(hosts, host+":"+hostAlias.IP)
+ }
+ }
+ p.HostAdd = hosts
+ }
+ podPorts := getPodPorts(podYAML.Spec.Containers)
+ p.PortMappings = podPorts
+
+ return p, nil
+}
+
+func ToSpecGen(ctx context.Context, containerYAML v1.Container, iid string, newImage *image.Image, volumes map[string]*KubeVolume, podID, podName, infraID string, configMaps []v1.ConfigMap, seccompPaths *KubeSeccompPaths, restartPolicy string) (*specgen.SpecGenerator, error) {
+ s := specgen.NewSpecGenerator(iid, false)
+
+ // podName should be non-empty for Deployment objects to be able to create
+ // multiple pods having containers with unique names
+ if len(podName) < 1 {
+ return nil, errors.Errorf("kubeContainerToCreateConfig got empty podName")
+ }
+
+ s.Name = fmt.Sprintf("%s-%s", podName, containerYAML.Name)
+
+ s.Terminal = containerYAML.TTY
+
+ s.Pod = podID
+
+ setupSecurityContext(s, containerYAML)
+
+ // Since we prefix the container name with pod name to work-around the uniqueness requirement,
+ // the seccomp profile should reference the actual container name from the YAML
+ // but apply to the containers with the prefixed name
+ s.SeccompProfilePath = seccompPaths.FindForContainer(containerYAML.Name)
+
+ s.ResourceLimits = &spec.LinuxResources{}
+ milliCPU, err := quantityToInt64(containerYAML.Resources.Limits.Cpu())
+ if err != nil {
+ return nil, errors.Wrap(err, "Failed to set CPU quota")
+ }
+ if milliCPU > 0 {
+ period, quota := util.CoresToPeriodAndQuota(float64(milliCPU) / 1000)
+ s.ResourceLimits.CPU = &spec.LinuxCPU{
+ Quota: &quota,
+ Period: &period,
+ }
+ }
+
+ limit, err := quantityToInt64(containerYAML.Resources.Limits.Memory())
+ if err != nil {
+ return nil, errors.Wrap(err, "Failed to set memory limit")
+ }
+
+ memoryRes, err := quantityToInt64(containerYAML.Resources.Requests.Memory())
+ if err != nil {
+ return nil, errors.Wrap(err, "Failed to set memory reservation")
+ }
+
+ if limit > 0 || memoryRes > 0 {
+ s.ResourceLimits.Memory = &spec.LinuxMemory{}
+ }
+
+ if limit > 0 {
+ s.ResourceLimits.Memory.Limit = &limit
+ }
+
+ if memoryRes > 0 {
+ s.ResourceLimits.Memory.Reservation = &memoryRes
+ }
+
+ // TODO: We dont understand why specgen does not take of this, but
+ // integration tests clearly pointed out that it was required.
+ s.Command = []string{}
+ imageData, err := newImage.Inspect(ctx)
+ if err != nil {
+ return nil, err
+ }
+ s.WorkDir = "/"
+ if imageData != nil && imageData.Config != nil {
+ if imageData.Config.WorkingDir != "" {
+ s.WorkDir = imageData.Config.WorkingDir
+ }
+ s.Command = imageData.Config.Entrypoint
+ s.Labels = imageData.Config.Labels
+ if len(imageData.Config.StopSignal) > 0 {
+ stopSignal, err := util.ParseSignal(imageData.Config.StopSignal)
+ if err != nil {
+ return nil, err
+ }
+ s.StopSignal = &stopSignal
+ }
+ }
+ if len(containerYAML.Command) != 0 {
+ s.Command = containerYAML.Command
+ }
+ // doc https://kubernetes.io/docs/tasks/inject-data-application/define-command-argument-container/#notes
+ if len(containerYAML.Args) != 0 {
+ s.Command = append(s.Command, containerYAML.Args...)
+ }
+ // FIXME,
+ // we are currently ignoring imageData.Config.ExposedPorts
+ if containerYAML.WorkingDir != "" {
+ s.WorkDir = containerYAML.WorkingDir
+ }
+
+ annotations := make(map[string]string)
+ if infraID != "" {
+ annotations[ann.SandboxID] = infraID
+ annotations[ann.ContainerType] = ann.ContainerTypeContainer
+ }
+ s.Annotations = annotations
+
+ // Environment Variables
+ envs := map[string]string{}
+ for _, env := range containerYAML.Env {
+ value := envVarValue(env, configMaps)
+
+ envs[env.Name] = value
+ }
+ for _, envFrom := range containerYAML.EnvFrom {
+ cmEnvs := envVarsFromConfigMap(envFrom, configMaps)
+
+ for k, v := range cmEnvs {
+ envs[k] = v
+ }
+ }
+ s.Env = envs
+
+ for _, volume := range containerYAML.VolumeMounts {
+ volumeSource, exists := volumes[volume.Name]
+ if !exists {
+ return nil, errors.Errorf("Volume mount %s specified for container but not configured in volumes", volume.Name)
+ }
+ switch volumeSource.Type {
+ case KubeVolumeTypeBindMount:
+ if err := parse.ValidateVolumeCtrDir(volume.MountPath); err != nil {
+ return nil, errors.Wrapf(err, "error in parsing MountPath")
+ }
+ mount := spec.Mount{
+ Destination: volume.MountPath,
+ Source: volumeSource.Source,
+ Type: "bind",
+ }
+ if volume.ReadOnly {
+ mount.Options = []string{"ro"}
+ }
+ s.Mounts = append(s.Mounts, mount)
+ case KubeVolumeTypeNamed:
+ namedVolume := specgen.NamedVolume{
+ Dest: volume.MountPath,
+ Name: volumeSource.Source,
+ }
+ if volume.ReadOnly {
+ namedVolume.Options = []string{"ro"}
+ }
+ s.Volumes = append(s.Volumes, &namedVolume)
+ default:
+ return nil, errors.Errorf("Unsupported volume source type")
+ }
+ }
+
+ s.RestartPolicy = restartPolicy
+
+ return s, nil
+}
+
+func setupSecurityContext(s *specgen.SpecGenerator, containerYAML v1.Container) {
+ if containerYAML.SecurityContext == nil {
+ return
+ }
+ if containerYAML.SecurityContext.ReadOnlyRootFilesystem != nil {
+ s.ReadOnlyFilesystem = *containerYAML.SecurityContext.ReadOnlyRootFilesystem
+ }
+ if containerYAML.SecurityContext.Privileged != nil {
+ s.Privileged = *containerYAML.SecurityContext.Privileged
+ }
+
+ if containerYAML.SecurityContext.AllowPrivilegeEscalation != nil {
+ s.NoNewPrivileges = !*containerYAML.SecurityContext.AllowPrivilegeEscalation
+ }
+
+ if seopt := containerYAML.SecurityContext.SELinuxOptions; seopt != nil {
+ if seopt.User != "" {
+ s.SelinuxOpts = append(s.SelinuxOpts, fmt.Sprintf("role:%s", seopt.User))
+ }
+ if seopt.Role != "" {
+ s.SelinuxOpts = append(s.SelinuxOpts, fmt.Sprintf("role:%s", seopt.Role))
+ }
+ if seopt.Type != "" {
+ s.SelinuxOpts = append(s.SelinuxOpts, fmt.Sprintf("role:%s", seopt.Type))
+ }
+ if seopt.Level != "" {
+ s.SelinuxOpts = append(s.SelinuxOpts, fmt.Sprintf("role:%s", seopt.Level))
+ }
+ }
+ if caps := containerYAML.SecurityContext.Capabilities; caps != nil {
+ for _, capability := range caps.Add {
+ s.CapAdd = append(s.CapAdd, string(capability))
+ }
+ for _, capability := range caps.Drop {
+ s.CapDrop = append(s.CapDrop, string(capability))
+ }
+ }
+ if containerYAML.SecurityContext.RunAsUser != nil {
+ s.User = fmt.Sprintf("%d", *containerYAML.SecurityContext.RunAsUser)
+ }
+ if containerYAML.SecurityContext.RunAsGroup != nil {
+ if s.User == "" {
+ s.User = "0"
+ }
+ s.User = fmt.Sprintf("%s:%d", s.User, *containerYAML.SecurityContext.RunAsGroup)
+ }
+}
+
+func quantityToInt64(quantity *resource.Quantity) (int64, error) {
+ if i, ok := quantity.AsInt64(); ok {
+ return i, nil
+ }
+
+ if i, ok := quantity.AsDec().Unscaled(); ok {
+ return i, nil
+ }
+
+ return 0, errors.Errorf("Quantity cannot be represented as int64: %v", quantity)
+}
+
+// envVarsFromConfigMap returns all key-value pairs as env vars from a configMap that matches the envFrom setting of a container
+func envVarsFromConfigMap(envFrom v1.EnvFromSource, configMaps []v1.ConfigMap) map[string]string {
+ envs := map[string]string{}
+
+ if envFrom.ConfigMapRef != nil {
+ cmName := envFrom.ConfigMapRef.Name
+
+ for _, c := range configMaps {
+ if cmName == c.Name {
+ envs = c.Data
+ break
+ }
+ }
+ }
+
+ return envs
+}
+
+// envVarValue returns the environment variable value configured within the container's env setting.
+// It gets the value from a configMap if specified, otherwise returns env.Value
+func envVarValue(env v1.EnvVar, configMaps []v1.ConfigMap) string {
+ for _, c := range configMaps {
+ if env.ValueFrom != nil {
+ if env.ValueFrom.ConfigMapKeyRef != nil {
+ if env.ValueFrom.ConfigMapKeyRef.Name == c.Name {
+ if value, ok := c.Data[env.ValueFrom.ConfigMapKeyRef.Key]; ok {
+ return value
+ }
+ }
+ }
+ }
+ }
+
+ return env.Value
+}
+
+// getPodPorts converts a slice of kube container descriptions to an
+// array of portmapping
+func getPodPorts(containers []v1.Container) []specgen.PortMapping {
+ var infraPorts []specgen.PortMapping
+ for _, container := range containers {
+ for _, p := range container.Ports {
+ if p.HostPort != 0 && p.ContainerPort == 0 {
+ p.ContainerPort = p.HostPort
+ }
+ if p.Protocol == "" {
+ p.Protocol = "tcp"
+ }
+ portBinding := specgen.PortMapping{
+ HostPort: uint16(p.HostPort),
+ ContainerPort: uint16(p.ContainerPort),
+ Protocol: strings.ToLower(string(p.Protocol)),
+ HostIP: p.HostIP,
+ }
+ // only hostPort is utilized in podman context, all container ports
+ // are accessible inside the shared network namespace
+ if p.HostPort != 0 {
+ infraPorts = append(infraPorts, portBinding)
+ }
+
+ }
+ }
+ return infraPorts
+}
diff --git a/pkg/specgen/generate/kube/play_test.go b/pkg/specgen/generate/kube/play_test.go
new file mode 100644
index 000000000..148540e9f
--- /dev/null
+++ b/pkg/specgen/generate/kube/play_test.go
@@ -0,0 +1,172 @@
+package kube
+
+import (
+ "testing"
+
+ "github.com/stretchr/testify/assert"
+ v1 "k8s.io/api/core/v1"
+ v12 "k8s.io/apimachinery/pkg/apis/meta/v1"
+)
+
+func TestEnvVarsFromConfigMap(t *testing.T) {
+ tests := []struct {
+ name string
+ envFrom v1.EnvFromSource
+ configMapList []v1.ConfigMap
+ expected map[string]string
+ }{
+ {
+ "ConfigMapExists",
+ v1.EnvFromSource{
+ ConfigMapRef: &v1.ConfigMapEnvSource{
+ LocalObjectReference: v1.LocalObjectReference{
+ Name: "foo",
+ },
+ },
+ },
+ configMapList,
+ map[string]string{
+ "myvar": "foo",
+ },
+ },
+ {
+ "ConfigMapDoesNotExist",
+ v1.EnvFromSource{
+ ConfigMapRef: &v1.ConfigMapEnvSource{
+ LocalObjectReference: v1.LocalObjectReference{
+ Name: "doesnotexist",
+ },
+ },
+ },
+ configMapList,
+ map[string]string{},
+ },
+ {
+ "EmptyConfigMapList",
+ v1.EnvFromSource{
+ ConfigMapRef: &v1.ConfigMapEnvSource{
+ LocalObjectReference: v1.LocalObjectReference{
+ Name: "foo",
+ },
+ },
+ },
+ []v1.ConfigMap{},
+ map[string]string{},
+ },
+ }
+
+ for _, test := range tests {
+ test := test
+ t.Run(test.name, func(t *testing.T) {
+ result := envVarsFromConfigMap(test.envFrom, test.configMapList)
+ assert.Equal(t, test.expected, result)
+ })
+ }
+}
+
+func TestEnvVarValue(t *testing.T) {
+ tests := []struct {
+ name string
+ envVar v1.EnvVar
+ configMapList []v1.ConfigMap
+ expected string
+ }{
+ {
+ "ConfigMapExists",
+ v1.EnvVar{
+ Name: "FOO",
+ ValueFrom: &v1.EnvVarSource{
+ ConfigMapKeyRef: &v1.ConfigMapKeySelector{
+ LocalObjectReference: v1.LocalObjectReference{
+ Name: "foo",
+ },
+ Key: "myvar",
+ },
+ },
+ },
+ configMapList,
+ "foo",
+ },
+ {
+ "ContainerKeyDoesNotExistInConfigMap",
+ v1.EnvVar{
+ Name: "FOO",
+ ValueFrom: &v1.EnvVarSource{
+ ConfigMapKeyRef: &v1.ConfigMapKeySelector{
+ LocalObjectReference: v1.LocalObjectReference{
+ Name: "foo",
+ },
+ Key: "doesnotexist",
+ },
+ },
+ },
+ configMapList,
+ "",
+ },
+ {
+ "ConfigMapDoesNotExist",
+ v1.EnvVar{
+ Name: "FOO",
+ ValueFrom: &v1.EnvVarSource{
+ ConfigMapKeyRef: &v1.ConfigMapKeySelector{
+ LocalObjectReference: v1.LocalObjectReference{
+ Name: "doesnotexist",
+ },
+ Key: "myvar",
+ },
+ },
+ },
+ configMapList,
+ "",
+ },
+ {
+ "EmptyConfigMapList",
+ v1.EnvVar{
+ Name: "FOO",
+ ValueFrom: &v1.EnvVarSource{
+ ConfigMapKeyRef: &v1.ConfigMapKeySelector{
+ LocalObjectReference: v1.LocalObjectReference{
+ Name: "foo",
+ },
+ Key: "myvar",
+ },
+ },
+ },
+ []v1.ConfigMap{},
+ "",
+ },
+ }
+
+ for _, test := range tests {
+ test := test
+ t.Run(test.name, func(t *testing.T) {
+ result := envVarValue(test.envVar, test.configMapList)
+ assert.Equal(t, test.expected, result)
+ })
+ }
+}
+
+var configMapList = []v1.ConfigMap{
+ {
+ TypeMeta: v12.TypeMeta{
+ Kind: "ConfigMap",
+ },
+ ObjectMeta: v12.ObjectMeta{
+ Name: "bar",
+ },
+ Data: map[string]string{
+ "myvar": "bar",
+ },
+ },
+ {
+ TypeMeta: v12.TypeMeta{
+ Kind: "ConfigMap",
+ },
+ ObjectMeta: v12.ObjectMeta{
+ Name: "foo",
+ },
+ Data: map[string]string{
+ "myvar": "foo",
+ },
+ },
+}
diff --git a/pkg/specgen/generate/kube/seccomp.go b/pkg/specgen/generate/kube/seccomp.go
new file mode 100644
index 000000000..4cbdf6e2e
--- /dev/null
+++ b/pkg/specgen/generate/kube/seccomp.go
@@ -0,0 +1,84 @@
+package kube
+
+import (
+ "path/filepath"
+ "strings"
+
+ "github.com/containers/podman/v2/libpod"
+ "github.com/pkg/errors"
+ v1 "k8s.io/api/core/v1"
+)
+
+// KubeSeccompPaths holds information about a pod YAML's seccomp configuration
+// it holds both container and pod seccomp paths
+type KubeSeccompPaths struct {
+ containerPaths map[string]string
+ podPath string
+}
+
+// FindForContainer checks whether a container has a seccomp path configured for it
+// if not, it returns the podPath, which should always have a value
+func (k *KubeSeccompPaths) FindForContainer(ctrName string) string {
+ if path, ok := k.containerPaths[ctrName]; ok {
+ return path
+ }
+ return k.podPath
+}
+
+// InitializeSeccompPaths takes annotations from the pod object metadata and finds annotations pertaining to seccomp
+// it parses both pod and container level
+// if the annotation is of the form "localhost/%s", the seccomp profile will be set to profileRoot/%s
+func InitializeSeccompPaths(annotations map[string]string, profileRoot string) (*KubeSeccompPaths, error) {
+ seccompPaths := &KubeSeccompPaths{containerPaths: make(map[string]string)}
+ var err error
+ if annotations != nil {
+ for annKeyValue, seccomp := range annotations {
+ // check if it is prefaced with container.seccomp.security.alpha.kubernetes.io/
+ prefixAndCtr := strings.Split(annKeyValue, "/")
+ if prefixAndCtr[0]+"/" != v1.SeccompContainerAnnotationKeyPrefix {
+ continue
+ } else if len(prefixAndCtr) != 2 {
+ // this could be caused by a user inputting either of
+ // container.seccomp.security.alpha.kubernetes.io{,/}
+ // both of which are invalid
+ return nil, errors.Errorf("Invalid seccomp path: %s", prefixAndCtr[0])
+ }
+
+ path, err := verifySeccompPath(seccomp, profileRoot)
+ if err != nil {
+ return nil, err
+ }
+ seccompPaths.containerPaths[prefixAndCtr[1]] = path
+ }
+
+ podSeccomp, ok := annotations[v1.SeccompPodAnnotationKey]
+ if ok {
+ seccompPaths.podPath, err = verifySeccompPath(podSeccomp, profileRoot)
+ } else {
+ seccompPaths.podPath, err = libpod.DefaultSeccompPath()
+ }
+ if err != nil {
+ return nil, err
+ }
+ }
+ return seccompPaths, nil
+}
+
+// verifySeccompPath takes a path and checks whether it is a default, unconfined, or a path
+// the available options are parsed as defined in https://kubernetes.io/docs/concepts/policy/pod-security-policy/#seccomp
+func verifySeccompPath(path string, profileRoot string) (string, error) {
+ switch path {
+ case v1.DeprecatedSeccompProfileDockerDefault:
+ fallthrough
+ case v1.SeccompProfileRuntimeDefault:
+ return libpod.DefaultSeccompPath()
+ case "unconfined":
+ return path, nil
+ default:
+ parts := strings.Split(path, "/")
+ if parts[0] == "localhost" {
+ return filepath.Join(profileRoot, parts[1]), nil
+ }
+ return "", errors.Errorf("invalid seccomp path: %s", path)
+ }
+}
diff --git a/pkg/specgen/generate/kube/volume.go b/pkg/specgen/generate/kube/volume.go
new file mode 100644
index 000000000..2ef0f4c23
--- /dev/null
+++ b/pkg/specgen/generate/kube/volume.go
@@ -0,0 +1,124 @@
+package kube
+
+import (
+ "os"
+
+ "github.com/containers/buildah/pkg/parse"
+ "github.com/containers/podman/v2/libpod"
+ "github.com/pkg/errors"
+ "github.com/sirupsen/logrus"
+ v1 "k8s.io/api/core/v1"
+)
+
+const (
+ // https://kubernetes.io/docs/concepts/storage/volumes/#hostpath
+ kubeDirectoryPermission = 0755
+ // https://kubernetes.io/docs/concepts/storage/volumes/#hostpath
+ kubeFilePermission = 0644
+)
+
+type KubeVolumeType int
+
+const (
+ KubeVolumeTypeBindMount KubeVolumeType = iota
+ KubeVolumeTypeNamed KubeVolumeType = iota
+)
+
+type KubeVolume struct {
+ // Type of volume to create
+ Type KubeVolumeType
+ // Path for bind mount or volume name for named volume
+ Source string
+}
+
+// Create a KubeVolume from an HostPathVolumeSource
+func VolumeFromHostPath(hostPath *v1.HostPathVolumeSource) (*KubeVolume, error) {
+ if hostPath.Type != nil {
+ switch *hostPath.Type {
+ case v1.HostPathDirectoryOrCreate:
+ if _, err := os.Stat(hostPath.Path); os.IsNotExist(err) {
+ if err := os.Mkdir(hostPath.Path, kubeDirectoryPermission); err != nil {
+ return nil, err
+ }
+ }
+ // Label a newly created volume
+ if err := libpod.LabelVolumePath(hostPath.Path); err != nil {
+ return nil, errors.Wrapf(err, "error giving %s a label", hostPath.Path)
+ }
+ case v1.HostPathFileOrCreate:
+ if _, err := os.Stat(hostPath.Path); os.IsNotExist(err) {
+ f, err := os.OpenFile(hostPath.Path, os.O_RDONLY|os.O_CREATE, kubeFilePermission)
+ if err != nil {
+ return nil, errors.Wrap(err, "error creating HostPath")
+ }
+ if err := f.Close(); err != nil {
+ logrus.Warnf("Error in closing newly created HostPath file: %v", err)
+ }
+ }
+ // unconditionally label a newly created volume
+ if err := libpod.LabelVolumePath(hostPath.Path); err != nil {
+ return nil, errors.Wrapf(err, "error giving %s a label", hostPath.Path)
+ }
+ case v1.HostPathSocket:
+ st, err := os.Stat(hostPath.Path)
+ if err != nil {
+ return nil, errors.Wrap(err, "error checking HostPathSocket")
+ }
+ if st.Mode()&os.ModeSocket != os.ModeSocket {
+ return nil, errors.Errorf("error checking HostPathSocket: path %s is not a socket", hostPath.Path)
+ }
+
+ case v1.HostPathDirectory:
+ case v1.HostPathFile:
+ case v1.HostPathUnset:
+ // do nothing here because we will verify the path exists in validateVolumeHostDir
+ break
+ default:
+ return nil, errors.Errorf("Invalid HostPath type %v", hostPath.Type)
+ }
+ }
+
+ if err := parse.ValidateVolumeHostDir(hostPath.Path); err != nil {
+ return nil, errors.Wrapf(err, "error in parsing HostPath in YAML")
+ }
+
+ return &KubeVolume{
+ Type: KubeVolumeTypeBindMount,
+ Source: hostPath.Path,
+ }, nil
+}
+
+// Create a KubeVolume from a PersistentVolumeClaimVolumeSource
+func VolumeFromPersistentVolumeClaim(claim *v1.PersistentVolumeClaimVolumeSource) (*KubeVolume, error) {
+ return &KubeVolume{
+ Type: KubeVolumeTypeNamed,
+ Source: claim.ClaimName,
+ }, nil
+}
+
+// Create a KubeVolume from one of the supported VolumeSource
+func VolumeFromSource(volumeSource v1.VolumeSource) (*KubeVolume, error) {
+ if volumeSource.HostPath != nil {
+ return VolumeFromHostPath(volumeSource.HostPath)
+ } else if volumeSource.PersistentVolumeClaim != nil {
+ return VolumeFromPersistentVolumeClaim(volumeSource.PersistentVolumeClaim)
+ } else {
+ return nil, errors.Errorf("HostPath and PersistentVolumeClaim are currently the conly supported VolumeSource")
+ }
+}
+
+// Create a map of volume name to KubeVolume
+func InitializeVolumes(specVolumes []v1.Volume) (map[string]*KubeVolume, error) {
+ volumes := make(map[string]*KubeVolume)
+
+ for _, specVolume := range specVolumes {
+ volume, err := VolumeFromSource(specVolume.VolumeSource)
+ if err != nil {
+ return nil, err
+ }
+
+ volumes[specVolume.Name] = volume
+ }
+
+ return volumes, nil
+}
diff --git a/pkg/specgen/generate/namespaces.go b/pkg/specgen/generate/namespaces.go
index 7e4f09dc4..ddc73ca61 100644
--- a/pkg/specgen/generate/namespaces.go
+++ b/pkg/specgen/generate/namespaces.go
@@ -127,6 +127,7 @@ func namespaceOptions(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.
return nil, errNoInfra
}
toReturn = append(toReturn, libpod.WithIPCNSFrom(infraCtr))
+ toReturn = append(toReturn, libpod.WithShmDir(infraCtr.ShmDir()))
case specgen.FromContainer:
ipcCtr, err := rt.LookupContainer(s.IpcNS.Value)
if err != nil {
@@ -278,7 +279,7 @@ func specConfigureNamespaces(s *specgen.SpecGenerator, g *generate.Generator, rt
switch s.PidNS.NSMode {
case specgen.Path:
if _, err := os.Stat(s.PidNS.Value); err != nil {
- return errors.Wrapf(err, "cannot find specified PID namespace path %q", s.PidNS.Value)
+ return errors.Wrap(err, "cannot find specified PID namespace path")
}
if err := g.AddOrReplaceLinuxNamespace(string(spec.PIDNamespace), s.PidNS.Value); err != nil {
return err
@@ -297,7 +298,7 @@ func specConfigureNamespaces(s *specgen.SpecGenerator, g *generate.Generator, rt
switch s.IpcNS.NSMode {
case specgen.Path:
if _, err := os.Stat(s.IpcNS.Value); err != nil {
- return errors.Wrapf(err, "cannot find specified IPC namespace path %q", s.IpcNS.Value)
+ return errors.Wrap(err, "cannot find specified IPC namespace path")
}
if err := g.AddOrReplaceLinuxNamespace(string(spec.IPCNamespace), s.IpcNS.Value); err != nil {
return err
@@ -316,7 +317,7 @@ func specConfigureNamespaces(s *specgen.SpecGenerator, g *generate.Generator, rt
switch s.UtsNS.NSMode {
case specgen.Path:
if _, err := os.Stat(s.UtsNS.Value); err != nil {
- return errors.Wrapf(err, "cannot find specified UTS namespace path %q", s.UtsNS.Value)
+ return errors.Wrap(err, "cannot find specified UTS namespace path")
}
if err := g.AddOrReplaceLinuxNamespace(string(spec.UTSNamespace), s.UtsNS.Value); err != nil {
return err
@@ -367,7 +368,7 @@ func specConfigureNamespaces(s *specgen.SpecGenerator, g *generate.Generator, rt
switch s.UserNS.NSMode {
case specgen.Path:
if _, err := os.Stat(s.UserNS.Value); err != nil {
- return errors.Wrapf(err, "cannot find specified user namespace path %s", s.UserNS.Value)
+ return errors.Wrap(err, "cannot find specified user namespace path")
}
if err := g.AddOrReplaceLinuxNamespace(string(spec.UserNamespace), s.UserNS.Value); err != nil {
return err
@@ -410,7 +411,7 @@ func specConfigureNamespaces(s *specgen.SpecGenerator, g *generate.Generator, rt
switch s.CgroupNS.NSMode {
case specgen.Path:
if _, err := os.Stat(s.CgroupNS.Value); err != nil {
- return errors.Wrapf(err, "cannot find specified cgroup namespace path %s", s.CgroupNS.Value)
+ return errors.Wrap(err, "cannot find specified cgroup namespace path")
}
if err := g.AddOrReplaceLinuxNamespace(string(spec.CgroupNamespace), s.CgroupNS.Value); err != nil {
return err
@@ -429,7 +430,7 @@ func specConfigureNamespaces(s *specgen.SpecGenerator, g *generate.Generator, rt
switch s.NetNS.NSMode {
case specgen.Path:
if _, err := os.Stat(s.NetNS.Value); err != nil {
- return errors.Wrapf(err, "cannot find specified network namespace path %s", s.NetNS.Value)
+ return errors.Wrap(err, "cannot find specified network namespace path")
}
if err := g.AddOrReplaceLinuxNamespace(string(spec.NetworkNamespace), s.NetNS.Value); err != nil {
return err
diff --git a/pkg/specgen/generate/security.go b/pkg/specgen/generate/security.go
index d17cd4a9a..dee140282 100644
--- a/pkg/specgen/generate/security.go
+++ b/pkg/specgen/generate/security.go
@@ -135,7 +135,9 @@ func securityConfigureGenerator(s *specgen.SpecGenerator, g *generate.Generator,
configSpec.Process.Capabilities.Bounding = caplist
configSpec.Process.Capabilities.Inheritable = caplist
- if s.User == "" || s.User == "root" || s.User == "0" {
+ user := strings.Split(s.User, ":")[0]
+
+ if (user == "" && s.UserNS.NSMode != specgen.KeepID) || user == "root" || user == "0" {
configSpec.Process.Capabilities.Effective = caplist
configSpec.Process.Capabilities.Permitted = caplist
} else {
@@ -145,6 +147,12 @@ func securityConfigureGenerator(s *specgen.SpecGenerator, g *generate.Generator,
}
configSpec.Process.Capabilities.Effective = userCaps
configSpec.Process.Capabilities.Permitted = userCaps
+
+ // Ambient capabilities were added to Linux 4.3. Set ambient
+ // capabilities only when the kernel supports them.
+ if supportAmbientCapabilities() {
+ configSpec.Process.Capabilities.Ambient = userCaps
+ }
}
g.SetProcessNoNewPrivileges(s.NoNewPrivileges)
diff --git a/pkg/specgen/generate/storage.go b/pkg/specgen/generate/storage.go
index b225f79ee..331a5c5bf 100644
--- a/pkg/specgen/generate/storage.go
+++ b/pkg/specgen/generate/storage.go
@@ -33,17 +33,17 @@ var (
)
// Produce final mounts and named volumes for a container
-func finalizeMounts(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.Runtime, rtc *config.Config, img *image.Image) ([]spec.Mount, []*specgen.NamedVolume, error) {
+func finalizeMounts(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.Runtime, rtc *config.Config, img *image.Image) ([]spec.Mount, []*specgen.NamedVolume, []*specgen.OverlayVolume, error) {
// Get image volumes
baseMounts, baseVolumes, err := getImageVolumes(ctx, img, s)
if err != nil {
- return nil, nil, err
+ return nil, nil, nil, err
}
// Get volumes-from mounts
volFromMounts, volFromVolumes, err := getVolumesFrom(s.VolumesFrom, rt)
if err != nil {
- return nil, nil, err
+ return nil, nil, nil, err
}
// Supersede from --volumes-from.
@@ -57,19 +57,53 @@ func finalizeMounts(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.Ru
// Need to make map forms of specgen mounts/volumes.
unifiedMounts := map[string]spec.Mount{}
unifiedVolumes := map[string]*specgen.NamedVolume{}
+ unifiedOverlays := map[string]*specgen.OverlayVolume{}
+
+ // Need to make map forms of specgen mounts/volumes.
+ commonMounts, commonVolumes, commonOverlayVolumes, err := specgen.GenVolumeMounts(rtc.Volumes())
+ if err != nil {
+ return nil, nil, nil, err
+ }
+
for _, m := range s.Mounts {
if _, ok := unifiedMounts[m.Destination]; ok {
- return nil, nil, errors.Wrapf(errDuplicateDest, "conflict in specified mounts - multiple mounts at %q", m.Destination)
+ return nil, nil, nil, errors.Wrapf(errDuplicateDest, "conflict in specified mounts - multiple mounts at %q", m.Destination)
}
unifiedMounts[m.Destination] = m
}
+
+ for _, m := range commonMounts {
+ if _, ok := unifiedMounts[m.Destination]; !ok {
+ unifiedMounts[m.Destination] = m
+ }
+ }
+
for _, v := range s.Volumes {
if _, ok := unifiedVolumes[v.Dest]; ok {
- return nil, nil, errors.Wrapf(errDuplicateDest, "conflict in specified volumes - multiple volumes at %q", v.Dest)
+ return nil, nil, nil, errors.Wrapf(errDuplicateDest, "conflict in specified volumes - multiple volumes at %q", v.Dest)
}
unifiedVolumes[v.Dest] = v
}
+ for _, v := range commonVolumes {
+ if _, ok := unifiedVolumes[v.Dest]; !ok {
+ unifiedVolumes[v.Dest] = v
+ }
+ }
+
+ for _, v := range s.OverlayVolumes {
+ if _, ok := unifiedOverlays[v.Destination]; ok {
+ return nil, nil, nil, errors.Wrapf(errDuplicateDest, "conflict in specified volumes - multiple volumes at %q", v.Destination)
+ }
+ unifiedOverlays[v.Destination] = v
+ }
+
+ for _, v := range commonOverlayVolumes {
+ if _, ok := unifiedOverlays[v.Destination]; ok {
+ unifiedOverlays[v.Destination] = v
+ }
+ }
+
// If requested, add container init binary
if s.Init {
initPath := s.InitPath
@@ -78,10 +112,10 @@ func finalizeMounts(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.Ru
}
initMount, err := addContainerInitBinary(s, initPath)
if err != nil {
- return nil, nil, err
+ return nil, nil, nil, err
}
if _, ok := unifiedMounts[initMount.Destination]; ok {
- return nil, nil, errors.Wrapf(errDuplicateDest, "conflict with mount added by --init to %q", initMount.Destination)
+ return nil, nil, nil, errors.Wrapf(errDuplicateDest, "conflict with mount added by --init to %q", initMount.Destination)
}
unifiedMounts[initMount.Destination] = initMount
}
@@ -115,12 +149,12 @@ func finalizeMounts(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.Ru
// Check for conflicts between named volumes and mounts
for dest := range baseMounts {
if _, ok := baseVolumes[dest]; ok {
- return nil, nil, errors.Wrapf(errDuplicateDest, "conflict at mount destination %v", dest)
+ return nil, nil, nil, errors.Wrapf(errDuplicateDest, "conflict at mount destination %v", dest)
}
}
for dest := range baseVolumes {
if _, ok := baseMounts[dest]; ok {
- return nil, nil, errors.Wrapf(errDuplicateDest, "conflict at mount destination %v", dest)
+ return nil, nil, nil, errors.Wrapf(errDuplicateDest, "conflict at mount destination %v", dest)
}
}
// Final step: maps to arrays
@@ -129,7 +163,7 @@ func finalizeMounts(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.Ru
if mount.Type == TypeBind {
absSrc, err := filepath.Abs(mount.Source)
if err != nil {
- return nil, nil, errors.Wrapf(err, "error getting absolute path of %s", mount.Source)
+ return nil, nil, nil, errors.Wrapf(err, "error getting absolute path of %s", mount.Source)
}
mount.Source = absSrc
}
@@ -140,7 +174,12 @@ func finalizeMounts(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.Ru
finalVolumes = append(finalVolumes, volume)
}
- return finalMounts, finalVolumes, nil
+ finalOverlays := make([]*specgen.OverlayVolume, 0, len(unifiedOverlays))
+ for _, volume := range unifiedOverlays {
+ finalOverlays = append(finalOverlays, volume)
+ }
+
+ return finalMounts, finalVolumes, finalOverlays, nil
}
// Get image volumes from the given image
diff --git a/pkg/specgen/generate/validate.go b/pkg/specgen/generate/validate.go
index ed337321b..f0ab4b994 100644
--- a/pkg/specgen/generate/validate.go
+++ b/pkg/specgen/generate/validate.go
@@ -1,22 +1,20 @@
package generate
import (
+ "os"
+ "path/filepath"
+
"github.com/containers/common/pkg/sysinfo"
"github.com/containers/podman/v2/pkg/cgroups"
"github.com/containers/podman/v2/pkg/specgen"
+ "github.com/containers/podman/v2/utils"
"github.com/pkg/errors"
)
-// Verify resource limits are sanely set, removing any limits that are not
-// possible with the current cgroups config.
-func verifyContainerResources(s *specgen.SpecGenerator) ([]string, error) {
+// Verify resource limits are sanely set when running on cgroup v1.
+func verifyContainerResourcesCgroupV1(s *specgen.SpecGenerator) ([]string, error) {
warnings := []string{}
- cgroup2, err := cgroups.IsCgroup2UnifiedMode()
- if err != nil || cgroup2 {
- return warnings, err
- }
-
sysInfo := sysinfo.New(true)
if s.ResourceLimits == nil {
@@ -24,9 +22,7 @@ func verifyContainerResources(s *specgen.SpecGenerator) ([]string, error) {
}
if s.ResourceLimits.Unified != nil {
- if !cgroup2 {
- return nil, errors.New("Cannot use --cgroup-conf without cgroup v2")
- }
+ return nil, errors.New("Cannot use --cgroup-conf without cgroup v2")
}
// Memory checks
@@ -38,7 +34,7 @@ func verifyContainerResources(s *specgen.SpecGenerator) ([]string, error) {
memory.Swap = nil
}
if memory.Limit != nil && memory.Swap != nil && !sysInfo.SwapLimit {
- warnings = append(warnings, "Your kernel does not support swap limit capabilities,or the cgroup is not mounted. Memory limited without swap.")
+ warnings = append(warnings, "Your kernel does not support swap limit capabilities or the cgroup is not mounted. Memory limited without swap.")
memory.Swap = nil
}
if memory.Limit != nil && memory.Swap != nil && *memory.Swap < *memory.Limit {
@@ -163,3 +159,48 @@ func verifyContainerResources(s *specgen.SpecGenerator) ([]string, error) {
return warnings, nil
}
+
+// Verify resource limits are sanely set when running on cgroup v2.
+func verifyContainerResourcesCgroupV2(s *specgen.SpecGenerator) ([]string, error) {
+ warnings := []string{}
+
+ if s.ResourceLimits == nil {
+ return warnings, nil
+ }
+
+ if s.ResourceLimits.Memory != nil && s.ResourceLimits.Memory.Swap != nil {
+ own, err := utils.GetOwnCgroup()
+ if err != nil {
+ return warnings, err
+ }
+ memoryMax := filepath.Join("/sys/fs/cgroup", own, "memory.max")
+ memorySwapMax := filepath.Join("/sys/fs/cgroup", own, "memory.swap.max")
+ _, errMemoryMax := os.Stat(memoryMax)
+ _, errMemorySwapMax := os.Stat(memorySwapMax)
+ // Differently than cgroup v1, the memory.*max files are not present in the
+ // root directory, so we cannot query directly that, so as best effort use
+ // the current cgroup.
+ // Check whether memory.max exists in the current cgroup and memory.swap.max
+ // does not. In this case we can be sure memory swap is not enabled.
+ // If both files don't exist, the memory controller might not be enabled
+ // for the current cgroup.
+ if errMemoryMax == nil && errMemorySwapMax != nil {
+ warnings = append(warnings, "Your kernel does not support swap limit capabilities or the cgroup is not mounted. Memory limited without swap.")
+ s.ResourceLimits.Memory.Swap = nil
+ }
+ }
+ return warnings, nil
+}
+
+// Verify resource limits are sanely set, removing any limits that are not
+// possible with the current cgroups config.
+func verifyContainerResources(s *specgen.SpecGenerator) ([]string, error) {
+ cgroup2, err := cgroups.IsCgroup2UnifiedMode()
+ if err != nil {
+ return []string{}, err
+ }
+ if cgroup2 {
+ return verifyContainerResourcesCgroupV2(s)
+ }
+ return verifyContainerResourcesCgroupV1(s)
+}
generated by cgit v1.2.3-54-g00ecf (git 2.39.0) at 2024-10-05 02:34:29 +0000