support DisableCgroup, DisableApparmor, RestrictOOMScoreAdj

Add following config for supporting "rootless" mode

* DisableCgroup: disable cgroup
* DisableApparmor: disable Apparmor
* RestrictOOMScoreAdj: restrict the lower bound of OOMScoreAdj

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

docs/config.md

@@ -44,6 +44,19 @@ The explanation and default value of each configuration item are as follows:
   # limit.
   max_container_log_line_size = 16384
 
+  # disable_cgroup indicates to disable the cgroup support.
+  # This is useful when the daemon does not have permission to access cgroup.
+  disable_cgroup = false
+
+  # disable_apparmor indicates to disable the apparmor support.
+  # This is useful when the daemon does not have permission to access apparmor.
+  disable_apparmor = false
+
+  # restrict_oom_score_adj indicates to limit the lower bound of OOMScoreAdj to
+  # the containerd's current OOMScoreAdj.
+  # This is useful when the containerd does not have permission to decrease OOMScoreAdj.
+  restrict_oom_score_adj = false
+
   # "plugins.cri.containerd" contains config related to containerd
   [plugins.cri.containerd]
 

pkg/config/config.go

@@ -142,6 +142,16 @@ type PluginConfig struct {
 	// Log line longer than the limit will be split into multiple lines. Non-positive
 	// value means no limit.
 	MaxContainerLogLineSize int `toml:"max_container_log_line_size" json:"maxContainerLogSize"`
+	// DisableCgroup indicates to disable the cgroup support.
+	// This is useful when the containerd does not have permission to access cgroup.
+	DisableCgroup bool `toml:"disable_cgroup" json:"disableCgroup"`
+	// DisableApparmor indicates to disable the apparmor support.
+	// This is useful when the containerd does not have permission to access Apparmor.
+	DisableApparmor bool `toml:"disable_apparmor" json:"disableApparmor"`
+	// RestrictOOMScoreAdj indicates to limit the lower bound of OOMScoreAdj to the containerd's
+	// current OOMScoreADj.
+	// This is useful when the containerd does not have permission to decrease OOMScoreAdj.
+	RestrictOOMScoreAdj bool `toml:"restrict_oom_score_adj" json:"restrictOOMScoreAdj"`
 }
 
 // X509KeyPairStreaming contains the x509 configuration for streaming

pkg/server/container_create.go

@@ -417,12 +417,18 @@ func (c *criService) generateContainerSpec(id string, sandboxID string, sandboxP
 
 	g.SetRootReadonly(securityContext.GetReadonlyRootfs())
 
-	setOCILinuxResource(&g, config.GetLinux().GetResources())
-
-	if sandboxConfig.GetLinux().GetCgroupParent() != "" {
-		cgroupsPath := getCgroupsPath(sandboxConfig.GetLinux().GetCgroupParent(), id,
-			c.config.SystemdCgroup)
-		g.SetLinuxCgroupsPath(cgroupsPath)
+	if c.config.DisableCgroup {
+		g.SetLinuxCgroupsPath("")
+	} else {
+		setOCILinuxResourceCgroup(&g, config.GetLinux().GetResources())
+		if sandboxConfig.GetLinux().GetCgroupParent() != "" {
+			cgroupsPath := getCgroupsPath(sandboxConfig.GetLinux().GetCgroupParent(), id,
+				c.config.SystemdCgroup)
+			g.SetLinuxCgroupsPath(cgroupsPath)
+		}
+	}
+	if err := setOCILinuxResourceOOMScoreAdj(&g, config.GetLinux().GetResources(), c.config.RestrictOOMScoreAdj); err != nil {
+		return nil, err
 	}
 
 	// Set namespaces, share namespace with sandbox container.
@@ -744,8 +750,8 @@ func setOCIBindMountsPrivileged(g *generate.Generator) {
 	spec.Linux.MaskedPaths = nil
 }
 
-// setOCILinuxResource set container resource limit.
-func setOCILinuxResource(g *generate.Generator, resources *runtime.LinuxContainerResources) {
+// setOCILinuxResourceCgroup set container cgroup resource limit.
+func setOCILinuxResourceCgroup(g *generate.Generator, resources *runtime.LinuxContainerResources) {
 	if resources == nil {
 		return
 	}
@@ -753,11 +759,28 @@ func setOCILinuxResource(g *generate.Generator, resources *runtime.LinuxContaine
 	g.SetLinuxResourcesCPUQuota(resources.GetCpuQuota())
 	g.SetLinuxResourcesCPUShares(uint64(resources.GetCpuShares()))
 	g.SetLinuxResourcesMemoryLimit(resources.GetMemoryLimitInBytes())
-	g.SetProcessOOMScoreAdj(int(resources.GetOomScoreAdj()))
 	g.SetLinuxResourcesCPUCpus(resources.GetCpusetCpus())
 	g.SetLinuxResourcesCPUMems(resources.GetCpusetMems())
 }
 
+// setOCILinuxResourceOOMScoreAdj set container OOMScoreAdj resource limit.
+func setOCILinuxResourceOOMScoreAdj(g *generate.Generator, resources *runtime.LinuxContainerResources, restrictOOMScoreAdjFlag bool) error {
+	if resources == nil {
+		return nil
+	}
+	adj := int(resources.GetOomScoreAdj())
+	if restrictOOMScoreAdjFlag {
+		var err error
+		adj, err = restrictOOMScoreAdj(adj)
+		if err != nil {
+			return err
+		}
+	}
+	g.SetProcessOOMScoreAdj(adj)
+
+	return nil
+}
+
 // getOCICapabilitiesList returns a list of all available capabilities.
 func getOCICapabilitiesList() []string {
 	var caps []string

pkg/server/helpers.go

@@ -18,6 +18,7 @@ package server
 
 import (
 	"fmt"
+	"io/ioutil"
 	"os"
 	"path"
 	"path/filepath"
@@ -496,3 +497,27 @@ func getRuntimeOptions(c containers.Container) (interface{}, error) {
 	}
 	return opts, nil
 }
+
+func getCurrentOOMScoreAdj() (int, error) {
+	b, err := ioutil.ReadFile("/proc/self/oom_score_adj")
+	if err != nil {
+		return 0, errors.Wrap(err, "could not get the daemon oom_score_adj")
+	}
+	s := strings.TrimSpace(string(b))
+	i, err := strconv.Atoi(s)
+	if err != nil {
+		return 0, errors.Wrap(err, "could not get the daemon oom_score_adj")
+	}
+	return i, nil
+}
+
+func restrictOOMScoreAdj(preferredOOMScoreAdj int) (int, error) {
+	currentOOMScoreAdj, err := getCurrentOOMScoreAdj()
+	if err != nil {
+		return preferredOOMScoreAdj, err
+	}
+	if preferredOOMScoreAdj < currentOOMScoreAdj {
+		return currentOOMScoreAdj, nil
+	}
+	return preferredOOMScoreAdj, nil
+}

pkg/server/sandbox_run.go

@@ -371,10 +371,14 @@ func (c *criService) generateSandboxContainerSpec(id string, config *runtime.Pod
 	// TODO(random-liu): [P2] Consider whether to add labels and annotations to the container.
 
 	// Set cgroups parent.
-	if config.GetLinux().GetCgroupParent() != "" {
-		cgroupsPath := getCgroupsPath(config.GetLinux().GetCgroupParent(), id,
-			c.config.SystemdCgroup)
-		g.SetLinuxCgroupsPath(cgroupsPath)
+	if c.config.DisableCgroup {
+		g.SetLinuxCgroupsPath("")
+	} else {
+		if config.GetLinux().GetCgroupParent() != "" {
+			cgroupsPath := getCgroupsPath(config.GetLinux().GetCgroupParent(), id,
+				c.config.SystemdCgroup)
+			g.SetLinuxCgroupsPath(cgroupsPath)
+		}
 	}
 	// When cgroup parent is not set, containerd-shim will create container in a child cgroup
 	// of the cgroup itself is in.
@@ -430,8 +434,17 @@ func (c *criService) generateSandboxContainerSpec(id string, config *runtime.Pod
 
 	// Note: LinuxSandboxSecurityContext does not currently provide an apparmor profile
 
-	g.SetLinuxResourcesCPUShares(uint64(defaultSandboxCPUshares))
-	g.SetProcessOOMScoreAdj(int(defaultSandboxOOMAdj))
+	if !c.config.DisableCgroup {
+		g.SetLinuxResourcesCPUShares(uint64(defaultSandboxCPUshares))
+	}
+	adj := int(defaultSandboxOOMAdj)
+	if c.config.RestrictOOMScoreAdj {
+		adj, err = restrictOOMScoreAdj(adj)
+		if err != nil {
+			return nil, err
+		}
+	}
+	g.SetProcessOOMScoreAdj(adj)
 
 	g.AddAnnotation(annotations.ContainerType, annotations.ContainerTypeSandbox)
 	g.AddAnnotation(annotations.SandboxID, id)

pkg/server/service.go

@@ -28,6 +28,7 @@ import (
 	cni "github.com/containerd/go-cni"
 	runcapparmor "github.com/opencontainers/runc/libcontainer/apparmor"
 	runcseccomp "github.com/opencontainers/runc/libcontainer/seccomp"
+	runcsystem "github.com/opencontainers/runc/libcontainer/system"
 	"github.com/opencontainers/selinux/go-selinux"
 	"github.com/pkg/errors"
 	"github.com/sirupsen/logrus"
@@ -108,7 +109,7 @@ func NewCRIService(config criconfig.Config, client *containerd.Client) (CRIServi
 	c := &criService{
 		config:             config,
 		client:             client,
-		apparmorEnabled:    runcapparmor.IsEnabled(),
+		apparmorEnabled:    runcapparmor.IsEnabled() && !config.DisableApparmor,
 		seccompEnabled:     runcseccomp.IsEnabled(),
 		os:                 osinterface.RealOS{},
 		sandboxStore:       sandboxstore.NewStore(),
@@ -120,6 +121,12 @@ func NewCRIService(config criconfig.Config, client *containerd.Client) (CRIServi
 		initialized:        atomic.NewBool(false),
 	}
 
+	if runcsystem.RunningInUserNS() {
+		if !(config.DisableCgroup && !c.apparmorEnabled && config.RestrictOOMScoreAdj) {
+			logrus.Warn("Running containerd in a user namespace typically requires disable_cgroup, disable_apparmor, restrict_oom_score_adj set to be true")
+		}
+	}
+
 	if c.config.EnableSelinux {
 		if !selinux.GetEnabled() {
 			logrus.Warn("Selinux is not supported")