diff --git a/pkg/sentry/fsimpl/cgroupfs/BUILD b/pkg/sentry/fsimpl/cgroupfs/BUILD
index fda2006ba..0e31bb719 100644
--- a/pkg/sentry/fsimpl/cgroupfs/BUILD
+++ b/pkg/sentry/fsimpl/cgroupfs/BUILD
@@ -40,6 +40,7 @@ go_library(
         "cpu.go",
         "cpuacct.go",
         "cpuset.go",
+        "devices.go",
         "dir_refs.go",
         "job.go",
         "memory.go",
diff --git a/pkg/sentry/fsimpl/cgroupfs/cgroupfs.go b/pkg/sentry/fsimpl/cgroupfs/cgroupfs.go
index e186c351d..3bce26e90 100644
--- a/pkg/sentry/fsimpl/cgroupfs/cgroupfs.go
+++ b/pkg/sentry/fsimpl/cgroupfs/cgroupfs.go
@@ -89,13 +89,14 @@ var allControllers = []kernel.CgroupControllerType{
 	kernel.CgroupControllerCPU,
 	kernel.CgroupControllerCPUAcct,
 	kernel.CgroupControllerCPUSet,
+	kernel.CgroupControllerDevices,
 	kernel.CgroupControllerJob,
 	kernel.CgroupControllerMemory,
 	kernel.CgroupControllerPIDs,
 }
 
 // SupportedMountOptions is the set of supported mount options for cgroupfs.
-var SupportedMountOptions = []string{"all", "cpu", "cpuacct", "cpuset", "job", "memory", "pids"}
+var SupportedMountOptions = []string{"all", "cpu", "cpuacct", "cpuset", "devices", "job", "memory", "pids"}
 
 // FilesystemType implements vfs.FilesystemType.
 //
@@ -227,6 +228,10 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
 		delete(mopts, "cpuset")
 		wantControllers = append(wantControllers, kernel.CgroupControllerCPUSet)
 	}
+	if _, ok := mopts["devices"]; ok {
+		delete(mopts, "devices")
+		wantControllers = append(wantControllers, kernel.CgroupControllerDevices)
+	}
 	if _, ok := mopts["job"]; ok {
 		delete(mopts, "job")
 		wantControllers = append(wantControllers, kernel.CgroupControllerJob)
@@ -341,6 +346,8 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
 			c = newCPUAcctController(fs)
 		case kernel.CgroupControllerCPUSet:
 			c = newCPUSetController(k, fs)
+		case kernel.CgroupControllerDevices:
+			c = newDevicesController(fs)
 		case kernel.CgroupControllerJob:
 			c = newJobController(fs)
 		case kernel.CgroupControllerMemory:
diff --git a/pkg/sentry/fsimpl/cgroupfs/devices.go b/pkg/sentry/fsimpl/cgroupfs/devices.go
new file mode 100644
index 000000000..c91e52b4c
--- /dev/null
+++ b/pkg/sentry/fsimpl/cgroupfs/devices.go
@@ -0,0 +1,163 @@
+// Copyright 2023 The gVisor Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package cgroupfs
+
+import (
+	"bytes"
+	"fmt"
+
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/sentry/fsimpl/kernfs"
+	"gvisor.dev/gvisor/pkg/sentry/kernel"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/vfs"
+	"gvisor.dev/gvisor/pkg/sync"
+	"gvisor.dev/gvisor/pkg/usermem"
+)
+
+const (
+	allowedDevices = "devices.allow"
+	deniedDevices  = "devices.deny"
+	wildcardDevice = 'a'
+)
+
+// permission represents a device access, read, write, and mknod.
+type permission string
+
+// +stateify savable
+type deviceRule struct {
+	// Device type, when the type is all, the following fields are ignored.
+	controllerType rune
+	// The device's major number.
+	major *int64
+	// The device's minor number.
+	minor *int64
+	// Cgroup access permission.
+	access permission
+}
+
+// +stateify savable
+type devicesController struct {
+	controllerCommon
+	controllerStateless
+	controllerNoResource
+
+	// mu protects the fields below.
+	mu sync.Mutex `state:"nosave"`
+
+	// Allow or deny the device rules below.
+	allow       bool
+	deviceRules []deviceRule
+}
+
+// +stateify savable
+type allowedDevicesData struct {
+	c *devicesController
+}
+
+// Generate implements vfs.DynamicBytesSource.Generate.
+func (d *allowedDevicesData) Generate(ctx context.Context, buf *bytes.Buffer) error {
+	return generate(ctx, buf, d.c, true)
+}
+
+// Write implements vfs.WritableDynamicBytesSource.Write.
+func (d *allowedDevicesData) Write(ctx context.Context, _ *vfs.FileDescription, src usermem.IOSequence, offset int64) (int64, error) {
+	return write(ctx, src, offset, d.c, true)
+}
+
+// +stateify savable
+type deniedDevicesData struct {
+	c *devicesController
+}
+
+// Generate implements vfs.DynamicBytesSource.Generate.
+func (d *deniedDevicesData) Generate(ctx context.Context, buf *bytes.Buffer) error {
+	return generate(ctx, buf, d.c, false)
+}
+
+// Write implements vfs.WritableDynamicBytesSource.Write.
+func (d *deniedDevicesData) Write(ctx context.Context, _ *vfs.FileDescription, src usermem.IOSequence, offset int64) (int64, error) {
+	return write(ctx, src, offset, d.c, true)
+}
+
+func generate(ctx context.Context, buf *bytes.Buffer, c *devicesController, allow bool) error {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	if allow == c.allow && len(c.deviceRules) > 0 {
+		for i, rule := range c.deviceRules {
+			if rule.controllerType == wildcardDevice {
+				buf.Reset()
+				buf.WriteRune(wildcardDevice)
+				return nil
+			}
+			buf.WriteString(deviceRuleString(rule))
+			if i < len(c.deviceRules)-1 {
+				buf.WriteRune(',')
+			}
+		}
+	} else {
+		buf.WriteString("")
+	}
+	return nil
+}
+
+func write(ctx context.Context, src usermem.IOSequence, offset int64, c *devicesController, allow bool) (int64, error) {
+	// TODO(b/289099718): add functions to add and remove rules when writing to device controller data.
+	return 0, nil
+}
+
+var _ controller = (*devicesController)(nil)
+
+func newDevicesController(fs *filesystem) *devicesController {
+	// The root device cgroup starts with rwm to all.
+	c := &devicesController{
+		allow:       true,
+		deviceRules: []deviceRule{},
+	}
+	c.controllerCommon.init(kernel.CgroupControllerJob, fs)
+	return c
+}
+
+// Clone implements controller.Clone.
+func (c *devicesController) Clone() controller {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	newRules := make([]deviceRule, len(c.deviceRules))
+	copy(newRules, c.deviceRules)
+	new := &devicesController{
+		allow:       c.allow,
+		deviceRules: newRules,
+	}
+	new.controllerCommon.cloneFromParent(c)
+	return new
+}
+
+// AddControlFiles implements controller.AddControlFiles.
+func (c *devicesController) AddControlFiles(ctx context.Context, creds *auth.Credentials, _ *cgroupInode, contents map[string]kernfs.Inode) {
+	contents[allowedDevices] = c.fs.newControllerWritableFile(ctx, creds, &allowedDevicesData{c: c}, true)
+	contents[deniedDevices] = c.fs.newControllerWritableFile(ctx, creds, &deniedDevicesData{c: c}, true)
+}
+
+func deviceRuleString(rule deviceRule) string {
+	return fmt.Sprintf("%c %s:%s %s", rule.controllerType, deviceNumber(rule.major), deviceNumber(rule.minor), rule.access)
+}
+
+// deviceNumber converts a device number to string.
+func deviceNumber(number *int64) string {
+	if number == nil {
+		return "*"
+	}
+	return fmt.Sprint(number)
+}
diff --git a/pkg/sentry/kernel/cgroup.go b/pkg/sentry/kernel/cgroup.go
index b7ab1963d..52f031d30 100644
--- a/pkg/sentry/kernel/cgroup.go
+++ b/pkg/sentry/kernel/cgroup.go
@@ -42,6 +42,7 @@ const (
 	CgroupControllerCPU     = CgroupControllerType("cpu")
 	CgroupControllerCPUAcct = CgroupControllerType("cpuacct")
 	CgroupControllerCPUSet  = CgroupControllerType("cpuset")
+	CgroupControllerDevices = CgroupControllerType("devices")
 	CgroupControllerJob     = CgroupControllerType("job")
 	CgroupControllerMemory  = CgroupControllerType("memory")
 	CgroupControllerPIDs    = CgroupControllerType("pids")
@@ -56,6 +57,8 @@ func ParseCgroupController(val string) (CgroupControllerType, error) {
 		return CgroupControllerCPUAcct, nil
 	case "cpuset":
 		return CgroupControllerCPUSet, nil
+	case "devices":
+		return CgroupControllerDevices, nil
 	case "job":
 		return CgroupControllerJob, nil
 	case "memory":
diff --git a/test/syscalls/linux/cgroup.cc b/test/syscalls/linux/cgroup.cc
index 8bd51ad97..db7ee4084 100644
--- a/test/syscalls/linux/cgroup.cc
+++ b/test/syscalls/linux/cgroup.cc
@@ -1428,6 +1428,17 @@ TEST(PIDsCgroup, RaceFSDestructionChargeUncharge) {
   });
 }
 
+TEST(DevicesCgroup, ControlFilesExist) {
+  SKIP_IF(!CgroupsAvailable());
+
+  Mounter m(ASSERT_NO_ERRNO_AND_VALUE(TempPath::CreateDir()));
+  Cgroup c = ASSERT_NO_ERRNO_AND_VALUE(m.MountCgroupfs("devices"));
+
+  // The root group starts with allowing rwm to all.
+  EXPECT_THAT(c.ReadControlFile("devices.allow"), IsPosixErrorOkAndHolds(""));
+  EXPECT_THAT(c.ReadControlFile("devices.deny"), IsPosixErrorOkAndHolds(""));
+}
+
 }  // namespace
 }  // namespace testing
 }  // namespace gvisor