Merge pull request #6108 from sudo-sturbia:msgqueue/syscalls

PiperOrigin-RevId: 386323389

12 files changed, 813 insertions, 330 deletions

diff --git a/pkg/sentry/kernel/BUILD b/pkg/sentry/kernel/BUILD
index 26614b029..c613f4932 100644
--- a/pkg/sentry/kernel/BUILD
+++ b/pkg/sentry/kernel/BUILD
@@ -256,6 +256,7 @@ go_library(
         "//pkg/sentry/kernel/auth",
         "//pkg/sentry/kernel/epoll",
         "//pkg/sentry/kernel/futex",
+        "//pkg/sentry/kernel/msgqueue",
         "//pkg/sentry/kernel/sched",
         "//pkg/sentry/kernel/semaphore",
         "//pkg/sentry/kernel/shm",
diff --git a/pkg/sentry/kernel/ipc/BUILD b/pkg/sentry/kernel/ipc/BUILD
new file mode 100644
index 000000000..e42a94e15
--- /dev/null
+++ b/pkg/sentry/kernel/ipc/BUILD
@@ -0,0 +1,20 @@
+load("//tools:defs.bzl", "go_library")
+
+package(licenses = ["notice"])
+
+go_library(
+    name = "ipc",
+    srcs = [
+        "object.go",
+        "registry.go",
+    ],
+    visibility = ["//pkg/sentry:internal"],
+    deps = [
+        "//pkg/abi/linux",
+        "//pkg/context",
+        "//pkg/errors/linuxerr",
+        "//pkg/log",
+        "//pkg/sentry/fs",
+        "//pkg/sentry/kernel/auth",
+    ],
+)
diff --git a/pkg/sentry/kernel/ipc/object.go b/pkg/sentry/kernel/ipc/object.go
new file mode 100644
index 000000000..387b35e7e
--- /dev/null
+++ b/pkg/sentry/kernel/ipc/object.go
@@ -0,0 +1,115 @@
+// Copyright 2021 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 ipc defines functionality and utilities common to sysvipc mechanisms.
+//
+// Lock ordering: [shm/semaphore/msgqueue].Registry.mu -> Mechanism
+package ipc
+
+import (
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+)
+
+// Key is a user-provided identifier for IPC objects.
+type Key int32
+
+// ID is a kernel identifier for IPC objects.
+type ID int32
+
+// Object represents an abstract IPC object with fields common to all IPC
+// mechanisms.
+//
+// +stateify savable
+type Object struct {
+	// User namespace which owns the IPC namespace which owns the IPC object.
+	// Immutable.
+	UserNS *auth.UserNamespace
+
+	// ID is a kernel identifier for the IPC object. Immutable.
+	ID ID
+
+	// Key is a user-provided identifier for the IPC object. Immutable.
+	Key Key
+
+	// Creator is the user who created the IPC object. Immutable.
+	Creator fs.FileOwner
+
+	// Owner is the current owner of the IPC object.
+	Owner fs.FileOwner
+
+	// Perms is the access permissions the IPC object.
+	Perms fs.FilePermissions
+}
+
+// Mechanism represents a SysV mechanism that holds an IPC object. It can also
+// be looked at as a container for an ipc.Object, which is by definition a fully
+// functional SysV object.
+type Mechanism interface {
+	// Lock behaves the same as Mutex.Lock on the mechanism.
+	Lock()
+
+	// Unlock behaves the same as Mutex.Unlock on the mechanism.
+	Unlock()
+
+	// Object returns a pointer to the mechanism's ipc.Object. Mechanism.Lock,
+	// and Mechanism.Unlock should be used when the object is used.
+	Object() *Object
+
+	// Destroy destroys the mechanism.
+	Destroy()
+}
+
+// NewObject returns a new, initialized ipc.Object. The newly returned object
+// doesn't have a valid ID. When the object is registered, the registry assigns
+// it a new unique ID.
+func NewObject(un *auth.UserNamespace, key Key, creator, owner fs.FileOwner, perms fs.FilePermissions) *Object {
+	return &Object{
+		UserNS:  un,
+		Key:     key,
+		Creator: creator,
+		Owner:   owner,
+		Perms:   perms,
+	}
+}
+
+// CheckOwnership verifies whether an IPC object may be accessed using creds as
+// an owner. See ipc/util.c:ipcctl_obtain_check() in Linux.
+func (o *Object) CheckOwnership(creds *auth.Credentials) bool {
+	if o.Owner.UID == creds.EffectiveKUID || o.Creator.UID == creds.EffectiveKUID {
+		return true
+	}
+
+	// Tasks with CAP_SYS_ADMIN may bypass ownership checks. Strangely, Linux
+	// doesn't use CAP_IPC_OWNER for this despite CAP_IPC_OWNER being documented
+	// for use to "override IPC ownership checks".
+	return creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, o.UserNS)
+}
+
+// CheckPermissions verifies whether an IPC object is accessible using creds for
+// access described by req. See ipc/util.c:ipcperms() in Linux.
+func (o *Object) CheckPermissions(creds *auth.Credentials, req fs.PermMask) bool {
+	p := o.Perms.Other
+	if o.Owner.UID == creds.EffectiveKUID {
+		p = o.Perms.User
+	} else if creds.InGroup(o.Owner.GID) {
+		p = o.Perms.Group
+	}
+
+	if p.SupersetOf(req) {
+		return true
+	}
+	return creds.HasCapabilityIn(linux.CAP_IPC_OWNER, o.UserNS)
+}
diff --git a/pkg/sentry/kernel/ipc/registry.go b/pkg/sentry/kernel/ipc/registry.go
new file mode 100644
index 000000000..91de19070
--- /dev/null
+++ b/pkg/sentry/kernel/ipc/registry.go
@@ -0,0 +1,196 @@
+// Copyright 2021 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 ipc
+
+import (
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/errors/linuxerr"
+	"gvisor.dev/gvisor/pkg/log"
+	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+)
+
+// Registry is similar to Object, but for registries. It represent an abstract
+// SysV IPC registry with fields common to all SysV registries. Registry is not
+// thread-safe, and should be protected using a mutex.
+//
+// +stateify savable
+type Registry struct {
+	// UserNS owning the IPC namespace this registry belongs to. Immutable.
+	UserNS *auth.UserNamespace
+
+	// objects is a map of IDs to IPC mechanisms.
+	objects map[ID]Mechanism
+
+	// KeysToIDs maps a lookup key to an ID.
+	keysToIDs map[Key]ID
+
+	// lastIDUsed is used to find the next available ID for object creation.
+	lastIDUsed ID
+}
+
+// NewRegistry return a new, initialized ipc.Registry.
+func NewRegistry(userNS *auth.UserNamespace) *Registry {
+	return &Registry{
+		UserNS:    userNS,
+		objects:   make(map[ID]Mechanism),
+		keysToIDs: make(map[Key]ID),
+	}
+}
+
+// Find uses key to search for and return a SysV mechanism. Find returns an
+// error if an object is found by shouldn't be, or if the user doesn't have
+// permission to use the object. If no object is found, Find checks create
+// flag, and returns an error only if it's false.
+func (r *Registry) Find(ctx context.Context, key Key, mode linux.FileMode, create, exclusive bool) (Mechanism, error) {
+	if id, ok := r.keysToIDs[key]; ok {
+		mech := r.objects[id]
+		mech.Lock()
+		defer mech.Unlock()
+
+		obj := mech.Object()
+		creds := auth.CredentialsFromContext(ctx)
+		if !obj.CheckPermissions(creds, fs.PermsFromMode(mode)) {
+			// The [calling process / user] does not have permission to access
+			// the set, and does not have the CAP_IPC_OWNER capability in the
+			// user namespace that governs its IPC namespace.
+			return nil, linuxerr.EACCES
+		}
+
+		if create && exclusive {
+			// IPC_CREAT and IPC_EXCL were specified, but an object already
+			// exists for key.
+			return nil, linuxerr.EEXIST
+		}
+		return mech, nil
+	}
+
+	if !create {
+		// No object exists for key and msgflg did not specify IPC_CREAT.
+		return nil, linuxerr.ENOENT
+	}
+
+	return nil, nil
+}
+
+// Register adds the given object into Registry.Objects, and assigns it a new
+// ID. It returns an error if all IDs are exhausted.
+func (r *Registry) Register(m Mechanism) error {
+	id, err := r.newID()
+	if err != nil {
+		return err
+	}
+
+	obj := m.Object()
+	obj.ID = id
+
+	r.objects[id] = m
+	r.keysToIDs[obj.Key] = id
+
+	return nil
+}
+
+// newID finds the first unused ID in the registry, and returns an error if
+// non is found.
+func (r *Registry) newID() (ID, error) {
+	// Find the next available ID.
+	for id := r.lastIDUsed + 1; id != r.lastIDUsed; id++ {
+		// Handle wrap around.
+		if id < 0 {
+			id = 0
+			continue
+		}
+		if r.objects[id] == nil {
+			r.lastIDUsed = id
+			return id, nil
+		}
+	}
+
+	log.Warningf("ids exhausted, they may be leaking")
+
+	// The man pages for shmget(2) mention that ENOSPC should be used if "All
+	// possible shared memory IDs have been taken (SHMMNI)". Other SysV
+	// mechanisms don't have a specific errno for running out of IDs, but they
+	// return ENOSPC if the max number of objects is exceeded, so we assume that
+	// it's the same case.
+	return 0, linuxerr.ENOSPC
+}
+
+// Remove removes the mechanism with the given id from the registry, and calls
+// mechanism.Destroy to perform mechanism-specific removal.
+func (r *Registry) Remove(id ID, creds *auth.Credentials) error {
+	mech := r.objects[id]
+	if mech == nil {
+		return linuxerr.EINVAL
+	}
+
+	mech.Lock()
+	defer mech.Unlock()
+
+	obj := mech.Object()
+
+	// The effective user ID of the calling process must match the creator or
+	// owner of the [mechanism], or the caller must be privileged.
+	if !obj.CheckOwnership(creds) {
+		return linuxerr.EPERM
+	}
+
+	delete(r.objects, obj.ID)
+	delete(r.keysToIDs, obj.Key)
+	mech.Destroy()
+
+	return nil
+}
+
+// ForAllObjects executes a given function for all given objects.
+func (r *Registry) ForAllObjects(f func(o Mechanism)) {
+	for _, o := range r.objects {
+		f(o)
+	}
+}
+
+// FindByID returns the mechanism with the given ID, nil if non exists.
+func (r *Registry) FindByID(id ID) Mechanism {
+	return r.objects[id]
+}
+
+// DissociateKey removes the association between a mechanism and its key
+// (deletes it from r.keysToIDs), preventing it from being discovered by any new
+// process, but not necessarily destroying it. If the given key doesn't exist,
+// nothing is changed.
+func (r *Registry) DissociateKey(key Key) {
+	delete(r.keysToIDs, key)
+}
+
+// DissociateID removes the association between a mechanism and its ID (deletes
+// it from r.objects). An ID can't be removed unless the associated key is
+// removed already, this is done to prevent the users from acquiring nil a
+// Mechanism.
+//
+// Precondition: must be preceded by a call to r.DissociateKey.
+func (r *Registry) DissociateID(id ID) {
+	delete(r.objects, id)
+}
+
+// ObjectCount returns the number of registered objects.
+func (r *Registry) ObjectCount() int {
+	return len(r.objects)
+}
+
+// LastIDUsed returns the last used ID.
+func (r *Registry) LastIDUsed() ID {
+	return r.lastIDUsed
+}
diff --git a/pkg/sentry/kernel/ipc_namespace.go b/pkg/sentry/kernel/ipc_namespace.go
index 9545bb5ef..0b101b1bb 100644
--- a/pkg/sentry/kernel/ipc_namespace.go
+++ b/pkg/sentry/kernel/ipc_namespace.go
@@ -17,6 +17,7 @@ package kernel
 import (
 	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/msgqueue"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/semaphore"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/shm"
 )
@@ -30,6 +31,7 @@ type IPCNamespace struct {
 	// User namespace which owns this IPC namespace. Immutable.
 	userNS *auth.UserNamespace
 
+	queues     *msgqueue.Registry
 	semaphores *semaphore.Registry
 	shms       *shm.Registry
 }
@@ -38,6 +40,7 @@ type IPCNamespace struct {
 func NewIPCNamespace(userNS *auth.UserNamespace) *IPCNamespace {
 	ns := &IPCNamespace{
 		userNS:     userNS,
+		queues:     msgqueue.NewRegistry(userNS),
 		semaphores: semaphore.NewRegistry(userNS),
 		shms:       shm.NewRegistry(userNS),
 	}
@@ -45,6 +48,11 @@ func NewIPCNamespace(userNS *auth.UserNamespace) *IPCNamespace {
 	return ns
 }
 
+// MsgqueueRegistry returns the message queue registry for this namespace.
+func (i *IPCNamespace) MsgqueueRegistry() *msgqueue.Registry {
+	return i.queues
+}
+
 // SemaphoreRegistry returns the semaphore set registry for this namespace.
 func (i *IPCNamespace) SemaphoreRegistry() *semaphore.Registry {
 	return i.semaphores
diff --git a/pkg/sentry/kernel/msgqueue/BUILD b/pkg/sentry/kernel/msgqueue/BUILD
new file mode 100644
index 000000000..5ec11e1f6
--- /dev/null
+++ b/pkg/sentry/kernel/msgqueue/BUILD
@@ -0,0 +1,36 @@
+load("//tools:defs.bzl", "go_library")
+load("//tools/go_generics:defs.bzl", "go_template_instance")
+
+package(licenses = ["notice"])
+
+go_template_instance(
+    name = "message_list",
+    out = "message_list.go",
+    package = "msgqueue",
+    prefix = "msg",
+    template = "//pkg/ilist:generic_list",
+    types = {
+        "Element": "*Message",
+        "Linker": "*Message",
+    },
+)
+
+go_library(
+    name = "msgqueue",
+    srcs = [
+        "message_list.go",
+        "msgqueue.go",
+    ],
+    visibility = ["//pkg/sentry:internal"],
+    deps = [
+        "//pkg/abi/linux",
+        "//pkg/context",
+        "//pkg/errors/linuxerr",
+        "//pkg/sentry/fs",
+        "//pkg/sentry/kernel/auth",
+        "//pkg/sentry/kernel/ipc",
+        "//pkg/sentry/kernel/time",
+        "//pkg/sync",
+        "//pkg/waiter",
+    ],
+)
diff --git a/pkg/sentry/kernel/msgqueue/msgqueue.go b/pkg/sentry/kernel/msgqueue/msgqueue.go
new file mode 100644
index 000000000..3ce926950
--- /dev/null
+++ b/pkg/sentry/kernel/msgqueue/msgqueue.go
@@ -0,0 +1,220 @@
+// Copyright 2021 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 msgqueue implements System V message queues.
+package msgqueue
+
+import (
+	"gvisor.dev/gvisor/pkg/abi/linux"
+	"gvisor.dev/gvisor/pkg/context"
+	"gvisor.dev/gvisor/pkg/errors/linuxerr"
+	"gvisor.dev/gvisor/pkg/sentry/fs"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/ipc"
+	ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
+	"gvisor.dev/gvisor/pkg/sync"
+	"gvisor.dev/gvisor/pkg/waiter"
+)
+
+const (
+	// System-wide limit for maximum number of queues.
+	maxQueues = linux.MSGMNI
+
+	// Maximum size of a queue in bytes.
+	maxQueueBytes = linux.MSGMNB
+
+	// Maximum size of a message in bytes.
+	maxMessageBytes = linux.MSGMAX
+)
+
+// Registry contains a set of message queues that can be referenced using keys
+// or IDs.
+//
+// +stateify savable
+type Registry struct {
+	// mu protects all the fields below.
+	mu sync.Mutex `state:"nosave"`
+
+	// reg defines basic fields and operations needed for all SysV registries.
+	reg *ipc.Registry
+}
+
+// NewRegistry returns a new Registry ready to be used.
+func NewRegistry(userNS *auth.UserNamespace) *Registry {
+	return &Registry{
+		reg: ipc.NewRegistry(userNS),
+	}
+}
+
+// Queue represents a SysV message queue, described by sysvipc(7).
+//
+// +stateify savable
+type Queue struct {
+	// registry is the registry owning this queue. Immutable.
+	registry *Registry
+
+	// mu protects all the fields below.
+	mu sync.Mutex `state:"nosave"`
+
+	// dead is set to true when a queue is removed from the registry and should
+	// not be used. Operations on the queue should check dead, and return
+	// EIDRM if set to true.
+	dead bool
+
+	// obj defines basic fields that should be included in all SysV IPC objects.
+	obj *ipc.Object
+
+	// senders holds a queue of blocked message senders. Senders are notified
+	// when enough space is available in the queue to insert their message.
+	senders waiter.Queue
+
+	// receivers holds a queue of blocked receivers. Receivers are notified
+	// when a new message is inserted into the queue and can be received.
+	receivers waiter.Queue
+
+	// messages is a list of sent messages.
+	messages msgList
+
+	// sendTime is the last time a msgsnd was perfomed.
+	sendTime ktime.Time
+
+	// receiveTime is the last time a msgrcv was performed.
+	receiveTime ktime.Time
+
+	// changeTime is the last time the queue was modified using msgctl.
+	changeTime ktime.Time
+
+	// byteCount is the current number of message bytes in the queue.
+	byteCount uint64
+
+	// messageCount is the current number of messages in the queue.
+	messageCount uint64
+
+	// maxBytes is the maximum allowed number of bytes in the queue, and is also
+	// used as a limit for the number of total possible messages.
+	maxBytes uint64
+
+	// sendPID is the PID of the process that performed the last msgsnd.
+	sendPID int32
+
+	// receivePID is the PID of the process that performed the last msgrcv.
+	receivePID int32
+}
+
+// Message represents a message exchanged through a Queue via msgsnd(2) and
+// msgrcv(2).
+//
+// +stateify savable
+type Message struct {
+	msgEntry
+
+	// mType is an integer representing the type of the sent message.
+	mType int64
+
+	// mText is an untyped block of memory.
+	mText []byte
+
+	// mSize is the size of mText.
+	mSize uint64
+}
+
+// FindOrCreate creates a new message queue or returns an existing one. See
+// msgget(2).
+func (r *Registry) FindOrCreate(ctx context.Context, key ipc.Key, mode linux.FileMode, private, create, exclusive bool) (*Queue, error) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+
+	if !private {
+		queue, err := r.reg.Find(ctx, key, mode, create, exclusive)
+		if err != nil {
+			return nil, err
+		}
+
+		if queue != nil {
+			return queue.(*Queue), nil
+		}
+	}
+
+	// Check system-wide limits.
+	if r.reg.ObjectCount() >= maxQueues {
+		return nil, linuxerr.ENOSPC
+	}
+
+	return r.newQueueLocked(ctx, key, fs.FileOwnerFromContext(ctx), fs.FilePermsFromMode(mode))
+}
+
+// newQueueLocked creates a new queue using the given fields. An error is
+// returned if there're no more available identifiers.
+//
+// Precondition: r.mu must be held.
+func (r *Registry) newQueueLocked(ctx context.Context, key ipc.Key, creator fs.FileOwner, perms fs.FilePermissions) (*Queue, error) {
+	q := &Queue{
+		registry:    r,
+		obj:         ipc.NewObject(r.reg.UserNS, key, creator, creator, perms),
+		sendTime:    ktime.ZeroTime,
+		receiveTime: ktime.ZeroTime,
+		changeTime:  ktime.NowFromContext(ctx),
+		maxBytes:    maxQueueBytes,
+	}
+
+	err := r.reg.Register(q)
+	if err != nil {
+		return nil, err
+	}
+	return q, nil
+}
+
+// Remove removes the queue with specified ID. All waiters (readers and
+// writers) and writers will be awakened and fail. Remove will return an error
+// if the ID is invalid, or the the user doesn't have privileges.
+func (r *Registry) Remove(id ipc.ID, creds *auth.Credentials) error {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+
+	r.reg.Remove(id, creds)
+	return nil
+}
+
+// Lock implements ipc.Mechanism.Lock.
+func (q *Queue) Lock() {
+	q.mu.Lock()
+}
+
+// Unlock implements ipc.mechanism.Unlock.
+//
+// +checklocksignore
+func (q *Queue) Unlock() {
+	q.mu.Unlock()
+}
+
+// Object implements ipc.Mechanism.Object.
+func (q *Queue) Object() *ipc.Object {
+	return q.obj
+}
+
+// Destroy implements ipc.Mechanism.Destroy.
+func (q *Queue) Destroy() {
+	q.dead = true
+
+	// Notify waiters. Senders and receivers will try to run, and return an
+	// error (EIDRM). Waiters should remove themselves from the queue after
+	// waking up.
+	q.senders.Notify(waiter.EventOut)
+	q.receivers.Notify(waiter.EventIn)
+}
+
+// ID returns queue's ID.
+func (q *Queue) ID() ipc.ID {
+	return q.obj.ID
+}
diff --git a/pkg/sentry/kernel/semaphore/BUILD b/pkg/sentry/kernel/semaphore/BUILD
index a787c00a8..2ae08ed12 100644
--- a/pkg/sentry/kernel/semaphore/BUILD
+++ b/pkg/sentry/kernel/semaphore/BUILD
@@ -26,9 +26,9 @@ go_library(
         "//pkg/abi/linux",
         "//pkg/context",
         "//pkg/errors/linuxerr",
-        "//pkg/log",
         "//pkg/sentry/fs",
         "//pkg/sentry/kernel/auth",
+        "//pkg/sentry/kernel/ipc",
         "//pkg/sentry/kernel/time",
         "//pkg/sync",
         "//pkg/syserror",
@@ -41,10 +41,11 @@ go_test(
     srcs = ["semaphore_test.go"],
     library = ":semaphore",
     deps = [
-        "//pkg/abi/linux",
-        "//pkg/context",
-        "//pkg/sentry/contexttest",
-        "//pkg/sentry/kernel/auth",
-        "//pkg/syserror",
+        "//pkg/abi/linux",  # keep
+        "//pkg/context",  # keep
+        "//pkg/sentry/contexttest",  # keep
+        "//pkg/sentry/kernel/auth",  # keep
+        "//pkg/sentry/kernel/ipc",  # keep
+        "//pkg/syserror",  # keep
     ],
 )
diff --git a/pkg/sentry/kernel/semaphore/semaphore.go b/pkg/sentry/kernel/semaphore/semaphore.go
index 485c3a788..b7879d284 100644
--- a/pkg/sentry/kernel/semaphore/semaphore.go
+++ b/pkg/sentry/kernel/semaphore/semaphore.go
@@ -21,9 +21,9 @@ import (
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/errors/linuxerr"
-	"gvisor.dev/gvisor/pkg/log"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/ipc"
 	ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
 	"gvisor.dev/gvisor/pkg/sync"
 	"gvisor.dev/gvisor/pkg/syserror"
@@ -47,15 +47,15 @@ const (
 //
 // +stateify savable
 type Registry struct {
-	// userNS owning the ipc name this registry belongs to. Immutable.
-	userNS *auth.UserNamespace
 	// mu protects all fields below.
-	mu         sync.Mutex `state:"nosave"`
-	semaphores map[int32]*Set
-	lastIDUsed int32
+	mu sync.Mutex `state:"nosave"`
+
+	// reg defines basic fields and operations needed for all SysV registries.
+	reg *ipc.Registry
+
 	// indexes maintains a mapping between a set's index in virtual array and
 	// its identifier.
-	indexes map[int32]int32
+	indexes map[int32]ipc.ID
 }
 
 // Set represents a set of semaphores that can be operated atomically.
@@ -65,19 +65,11 @@ type Set struct {
 	// registry owning this sem set. Immutable.
 	registry *Registry
 
-	// Id is a handle that identifies the set.
-	ID int32
-
-	// key is an user provided key that can be shared between processes.
-	key int32
+	// mu protects all fields below.
+	mu sync.Mutex `state:"nosave"`
 
-	// creator is the user that created the set. Immutable.
-	creator fs.FileOwner
+	obj *ipc.Object
 
-	// mu protects all fields below.
-	mu         sync.Mutex `state:"nosave"`
-	owner      fs.FileOwner
-	perms      fs.FilePermissions
 	opTime     ktime.Time
 	changeTime ktime.Time
 
@@ -115,9 +107,8 @@ type waiter struct {
 // NewRegistry creates a new semaphore set registry.
 func NewRegistry(userNS *auth.UserNamespace) *Registry {
 	return &Registry{
-		userNS:     userNS,
-		semaphores: make(map[int32]*Set),
-		indexes:    make(map[int32]int32),
+		reg:     ipc.NewRegistry(userNS),
+		indexes: make(map[int32]ipc.ID),
 	}
 }
 
@@ -126,7 +117,7 @@ func NewRegistry(userNS *auth.UserNamespace) *Registry {
 // a new set is always created. If create is false, it fails if a set cannot
 // be found. If exclusive is true, it fails if a set with the same key already
 // exists.
-func (r *Registry) FindOrCreate(ctx context.Context, key, nsems int32, mode linux.FileMode, private, create, exclusive bool) (*Set, error) {
+func (r *Registry) FindOrCreate(ctx context.Context, key ipc.Key, nsems int32, mode linux.FileMode, private, create, exclusive bool) (*Set, error) {
 	if nsems < 0 || nsems > semsMax {
 		return nil, linuxerr.EINVAL
 	}
@@ -135,31 +126,19 @@ func (r *Registry) FindOrCreate(ctx context.Context, key, nsems int32, mode linu
 	defer r.mu.Unlock()
 
 	if !private {
-		// Look up an existing semaphore.
-		if set := r.findByKey(key); set != nil {
-			set.mu.Lock()
-			defer set.mu.Unlock()
-
-			// Check that caller can access semaphore set.
-			creds := auth.CredentialsFromContext(ctx)
-			if !set.checkPerms(creds, fs.PermsFromMode(mode)) {
-				return nil, linuxerr.EACCES
-			}
+		set, err := r.reg.Find(ctx, key, mode, create, exclusive)
+		if err != nil {
+			return nil, err
+		}
 
-			// Validate parameters.
+		// Validate semaphore-specific parameters.
+		if set != nil {
+			set := set.(*Set)
 			if nsems > int32(set.Size()) {
 				return nil, linuxerr.EINVAL
 			}
-			if create && exclusive {
-				return nil, linuxerr.EEXIST
-			}
 			return set, nil
 		}
-
-		if !create {
-			// Semaphore not found and should not be created.
-			return nil, syserror.ENOENT
-		}
 	}
 
 	// Zero is only valid if an existing set is found.
@@ -169,9 +148,9 @@ func (r *Registry) FindOrCreate(ctx context.Context, key, nsems int32, mode linu
 
 	// Apply system limits.
 	//
-	// Map semaphores and map indexes in a registry are of the same size,
-	// check map semaphores only here for the system limit.
-	if len(r.semaphores) >= setsMax {
+	// Map reg.objects and map indexes in a registry are of the same size,
+	// check map reg.objects only here for the system limit.
+	if r.reg.ObjectCount() >= setsMax {
 		return nil, syserror.ENOSPC
 	}
 	if r.totalSems() > int(semsTotalMax-nsems) {
@@ -179,9 +158,7 @@ func (r *Registry) FindOrCreate(ctx context.Context, key, nsems int32, mode linu
 	}
 
 	// Finally create a new set.
-	owner := fs.FileOwnerFromContext(ctx)
-	perms := fs.FilePermsFromMode(mode)
-	return r.newSet(ctx, key, owner, owner, perms, nsems)
+	return r.newSetLocked(ctx, key, fs.FileOwnerFromContext(ctx), fs.FilePermsFromMode(mode), nsems)
 }
 
 // IPCInfo returns information about system-wide semaphore limits and parameters.
@@ -208,7 +185,7 @@ func (r *Registry) SemInfo() *linux.SemInfo {
 	defer r.mu.Unlock()
 
 	info := r.IPCInfo()
-	info.SemUsz = uint32(len(r.semaphores))
+	info.SemUsz = uint32(r.reg.ObjectCount())
 	info.SemAem = uint32(r.totalSems())
 
 	return info
@@ -231,77 +208,59 @@ func (r *Registry) HighestIndex() int32 {
 	return highestIndex
 }
 
-// RemoveID removes set with give 'id' from the registry and marks the set as
+// Remove removes set with give 'id' from the registry and marks the set as
 // dead. All waiters will be awakened and fail.
-func (r *Registry) RemoveID(id int32, creds *auth.Credentials) error {
+func (r *Registry) Remove(id ipc.ID, creds *auth.Credentials) error {
 	r.mu.Lock()
 	defer r.mu.Unlock()
 
-	set := r.semaphores[id]
-	if set == nil {
-		return linuxerr.EINVAL
-	}
+	r.reg.Remove(id, creds)
+
 	index, found := r.findIndexByID(id)
 	if !found {
 		// Inconsistent state.
 		panic(fmt.Sprintf("unable to find an index for ID: %d", id))
 	}
-
-	set.mu.Lock()
-	defer set.mu.Unlock()
-
-	// "The effective user ID of the calling process must match the creator or
-	// owner of the semaphore set, or the caller must be privileged."
-	if !set.checkCredentials(creds) && !set.checkCapability(creds) {
-		return linuxerr.EACCES
-	}
-
-	delete(r.semaphores, set.ID)
 	delete(r.indexes, index)
-	set.destroy()
+
 	return nil
 }
 
-func (r *Registry) newSet(ctx context.Context, key int32, owner, creator fs.FileOwner, perms fs.FilePermissions, nsems int32) (*Set, error) {
+// newSetLocked creates a new Set using given fields. An error is returned if there
+// are no more available identifiers.
+//
+// Precondition: r.mu must be held.
+func (r *Registry) newSetLocked(ctx context.Context, key ipc.Key, creator fs.FileOwner, perms fs.FilePermissions, nsems int32) (*Set, error) {
 	set := &Set{
 		registry:   r,
-		key:        key,
-		owner:      owner,
-		creator:    owner,
-		perms:      perms,
+		obj:        ipc.NewObject(r.reg.UserNS, ipc.Key(key), creator, creator, perms),
 		changeTime: ktime.NowFromContext(ctx),
 		sems:       make([]sem, nsems),
 	}
 
-	// Find the next available ID.
-	for id := r.lastIDUsed + 1; id != r.lastIDUsed; id++ {
-		// Handle wrap around.
-		if id < 0 {
-			id = 0
-			continue
-		}
-		if r.semaphores[id] == nil {
-			index, found := r.findFirstAvailableIndex()
-			if !found {
-				panic("unable to find an available index")
-			}
-			r.indexes[index] = id
-			r.lastIDUsed = id
-			r.semaphores[id] = set
-			set.ID = id
-			return set, nil
-		}
+	err := r.reg.Register(set)
+	if err != nil {
+		return nil, err
 	}
 
-	log.Warningf("Semaphore map is full, they must be leaking")
-	return nil, syserror.ENOMEM
+	index, found := r.findFirstAvailableIndex()
+	if !found {
+		panic("unable to find an available index")
+	}
+	r.indexes[index] = set.obj.ID
+
+	return set, nil
 }
 
 // FindByID looks up a set given an ID.
-func (r *Registry) FindByID(id int32) *Set {
+func (r *Registry) FindByID(id ipc.ID) *Set {
 	r.mu.Lock()
 	defer r.mu.Unlock()
-	return r.semaphores[id]
+	mech := r.reg.FindByID(id)
+	if mech == nil {
+		return nil
+	}
+	return mech.(*Set)
 }
 
 // FindByIndex looks up a set given an index.
@@ -313,19 +272,10 @@ func (r *Registry) FindByIndex(index int32) *Set {
 	if !present {
 		return nil
 	}
-	return r.semaphores[id]
+	return r.reg.FindByID(id).(*Set)
 }
 
-func (r *Registry) findByKey(key int32) *Set {
-	for _, v := range r.semaphores {
-		if v.key == key {
-			return v
-		}
-	}
-	return nil
-}
-
-func (r *Registry) findIndexByID(id int32) (int32, bool) {
+func (r *Registry) findIndexByID(id ipc.ID) (int32, bool) {
 	for k, v := range r.indexes {
 		if v == id {
 			return k, true
@@ -345,12 +295,36 @@ func (r *Registry) findFirstAvailableIndex() (int32, bool) {
 
 func (r *Registry) totalSems() int {
 	totalSems := 0
-	for _, v := range r.semaphores {
-		totalSems += v.Size()
-	}
+	r.reg.ForAllObjects(
+		func(o ipc.Mechanism) {
+			totalSems += o.(*Set).Size()
+		},
+	)
 	return totalSems
 }
 
+// ID returns semaphore's ID.
+func (s *Set) ID() ipc.ID {
+	return s.obj.ID
+}
+
+// Object implements ipc.Mechanism.Object.
+func (s *Set) Object() *ipc.Object {
+	return s.obj
+}
+
+// Lock implements ipc.Mechanism.Lock.
+func (s *Set) Lock() {
+	s.mu.Lock()
+}
+
+// Unlock implements ipc.mechanism.Unlock.
+//
+// +checklocksignore
+func (s *Set) Unlock() {
+	s.mu.Unlock()
+}
+
 func (s *Set) findSem(num int32) *sem {
 	if num < 0 || int(num) >= s.Size() {
 		return nil
@@ -370,12 +344,12 @@ func (s *Set) Change(ctx context.Context, creds *auth.Credentials, owner fs.File
 
 	// "The effective UID of the calling process must match the owner or creator
 	// of the semaphore set, or the caller must be privileged."
-	if !s.checkCredentials(creds) && !s.checkCapability(creds) {
+	if !s.obj.CheckOwnership(creds) {
 		return linuxerr.EACCES
 	}
 
-	s.owner = owner
-	s.perms = perms
+	s.obj.Owner = owner
+	s.obj.Perms = perms
 	s.changeTime = ktime.NowFromContext(ctx)
 	return nil
 }
@@ -395,18 +369,18 @@ func (s *Set) semStat(creds *auth.Credentials, permMask fs.PermMask) (*linux.Sem
 	s.mu.Lock()
 	defer s.mu.Unlock()
 
-	if !s.checkPerms(creds, permMask) {
+	if !s.obj.CheckPermissions(creds, permMask) {
 		return nil, linuxerr.EACCES
 	}
 
 	return &linux.SemidDS{
 		SemPerm: linux.IPCPerm{
-			Key:  uint32(s.key),
-			UID:  uint32(creds.UserNamespace.MapFromKUID(s.owner.UID)),
-			GID:  uint32(creds.UserNamespace.MapFromKGID(s.owner.GID)),
-			CUID: uint32(creds.UserNamespace.MapFromKUID(s.creator.UID)),
-			CGID: uint32(creds.UserNamespace.MapFromKGID(s.creator.GID)),
-			Mode: uint16(s.perms.LinuxMode()),
+			Key:  uint32(s.obj.Key),
+			UID:  uint32(creds.UserNamespace.MapFromKUID(s.obj.Owner.UID)),
+			GID:  uint32(creds.UserNamespace.MapFromKGID(s.obj.Owner.GID)),
+			CUID: uint32(creds.UserNamespace.MapFromKUID(s.obj.Creator.UID)),
+			CGID: uint32(creds.UserNamespace.MapFromKGID(s.obj.Creator.GID)),
+			Mode: uint16(s.obj.Perms.LinuxMode()),
 			Seq:  0, // IPC sequence not supported.
 		},
 		SemOTime: s.opTime.TimeT(),
@@ -425,7 +399,7 @@ func (s *Set) SetVal(ctx context.Context, num int32, val int16, creds *auth.Cred
 	defer s.mu.Unlock()
 
 	// "The calling process must have alter permission on the semaphore set."
-	if !s.checkPerms(creds, fs.PermMask{Write: true}) {
+	if !s.obj.CheckPermissions(creds, fs.PermMask{Write: true}) {
 		return linuxerr.EACCES
 	}
 
@@ -461,7 +435,7 @@ func (s *Set) SetValAll(ctx context.Context, vals []uint16, creds *auth.Credenti
 	defer s.mu.Unlock()
 
 	// "The calling process must have alter permission on the semaphore set."
-	if !s.checkPerms(creds, fs.PermMask{Write: true}) {
+	if !s.obj.CheckPermissions(creds, fs.PermMask{Write: true}) {
 		return linuxerr.EACCES
 	}
 
@@ -483,7 +457,7 @@ func (s *Set) GetVal(num int32, creds *auth.Credentials) (int16, error) {
 	defer s.mu.Unlock()
 
 	// "The calling process must have read permission on the semaphore set."
-	if !s.checkPerms(creds, fs.PermMask{Read: true}) {
+	if !s.obj.CheckPermissions(creds, fs.PermMask{Read: true}) {
 		return 0, linuxerr.EACCES
 	}
 
@@ -500,7 +474,7 @@ func (s *Set) GetValAll(creds *auth.Credentials) ([]uint16, error) {
 	defer s.mu.Unlock()
 
 	// "The calling process must have read permission on the semaphore set."
-	if !s.checkPerms(creds, fs.PermMask{Read: true}) {
+	if !s.obj.CheckPermissions(creds, fs.PermMask{Read: true}) {
 		return nil, linuxerr.EACCES
 	}
 
@@ -517,7 +491,7 @@ func (s *Set) GetPID(num int32, creds *auth.Credentials) (int32, error) {
 	defer s.mu.Unlock()
 
 	// "The calling process must have read permission on the semaphore set."
-	if !s.checkPerms(creds, fs.PermMask{Read: true}) {
+	if !s.obj.CheckPermissions(creds, fs.PermMask{Read: true}) {
 		return 0, linuxerr.EACCES
 	}
 
@@ -533,7 +507,7 @@ func (s *Set) countWaiters(num int32, creds *auth.Credentials, pred func(w *wait
 	defer s.mu.Unlock()
 
 	// The calling process must have read permission on the semaphore set.
-	if !s.checkPerms(creds, fs.PermMask{Read: true}) {
+	if !s.obj.CheckPermissions(creds, fs.PermMask{Read: true}) {
 		return 0, linuxerr.EACCES
 	}
 
@@ -589,7 +563,7 @@ func (s *Set) ExecuteOps(ctx context.Context, ops []linux.Sembuf, creds *auth.Cr
 		}
 	}
 
-	if !s.checkPerms(creds, fs.PermMask{Read: readOnly, Write: !readOnly}) {
+	if !s.obj.CheckPermissions(creds, fs.PermMask{Read: readOnly, Write: !readOnly}) {
 		return nil, 0, linuxerr.EACCES
 	}
 
@@ -675,38 +649,10 @@ func (s *Set) AbortWait(num int32, ch chan struct{}) {
 	// Waiter may not be found in case it raced with wakeWaiters().
 }
 
-func (s *Set) checkCredentials(creds *auth.Credentials) bool {
-	return s.owner.UID == creds.EffectiveKUID ||
-		s.owner.GID == creds.EffectiveKGID ||
-		s.creator.UID == creds.EffectiveKUID ||
-		s.creator.GID == creds.EffectiveKGID
-}
-
-func (s *Set) checkCapability(creds *auth.Credentials) bool {
-	return creds.HasCapabilityIn(linux.CAP_IPC_OWNER, s.registry.userNS) && creds.UserNamespace.MapFromKUID(s.owner.UID).Ok()
-}
-
-func (s *Set) checkPerms(creds *auth.Credentials, reqPerms fs.PermMask) bool {
-	// Are we owner, or in group, or other?
-	p := s.perms.Other
-	if s.owner.UID == creds.EffectiveKUID {
-		p = s.perms.User
-	} else if creds.InGroup(s.owner.GID) {
-		p = s.perms.Group
-	}
-
-	// Are permissions satisfied without capability checks?
-	if p.SupersetOf(reqPerms) {
-		return true
-	}
-
-	return s.checkCapability(creds)
-}
-
-// destroy destroys the set.
+// Destroy implements ipc.Mechanism.Destroy.
 //
 // Preconditions: Caller must hold 's.mu'.
-func (s *Set) destroy() {
+func (s *Set) Destroy() {
 	// Notify all waiters. They will fail on the next attempt to execute
 	// operations and return error.
 	s.dead = true
diff --git a/pkg/sentry/kernel/semaphore/semaphore_test.go b/pkg/sentry/kernel/semaphore/semaphore_test.go
index e47acefdf..2e4ab8121 100644
--- a/pkg/sentry/kernel/semaphore/semaphore_test.go
+++ b/pkg/sentry/kernel/semaphore/semaphore_test.go
@@ -21,6 +21,7 @@ import (
 	"gvisor.dev/gvisor/pkg/context"
 	"gvisor.dev/gvisor/pkg/sentry/contexttest"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/ipc"
 	"gvisor.dev/gvisor/pkg/syserror"
 )
 
@@ -55,7 +56,7 @@ func signalled(ch chan struct{}) bool {
 
 func TestBasic(t *testing.T) {
 	ctx := contexttest.Context(t)
-	set := &Set{ID: 123, sems: make([]sem, 1)}
+	set := &Set{obj: &ipc.Object{ID: 123}, sems: make([]sem, 1)}
 	ops := []linux.Sembuf{
 		{SemOp: 1},
 	}
@@ -76,7 +77,7 @@ func TestBasic(t *testing.T) {
 
 func TestWaitForZero(t *testing.T) {
 	ctx := contexttest.Context(t)
-	set := &Set{ID: 123, sems: make([]sem, 1)}
+	set := &Set{obj: &ipc.Object{ID: 123}, sems: make([]sem, 1)}
 	ops := []linux.Sembuf{
 		{SemOp: 0},
 	}
@@ -115,7 +116,7 @@ func TestWaitForZero(t *testing.T) {
 
 func TestNoWait(t *testing.T) {
 	ctx := contexttest.Context(t)
-	set := &Set{ID: 123, sems: make([]sem, 1)}
+	set := &Set{obj: &ipc.Object{ID: 123}, sems: make([]sem, 1)}
 	ops := []linux.Sembuf{
 		{SemOp: 1},
 	}
@@ -138,11 +139,12 @@ func TestUnregister(t *testing.T) {
 	ctx := contexttest.Context(t)
 	r := NewRegistry(auth.NewRootUserNamespace())
 	set, err := r.FindOrCreate(ctx, 123, 2, linux.FileMode(0x600), true, true, true)
+
 	if err != nil {
 		t.Fatalf("FindOrCreate() failed, err: %v", err)
 	}
-	if got := r.FindByID(set.ID); got.ID != set.ID {
-		t.Fatalf("FindById(%d) failed, got: %+v, expected: %+v", set.ID, got, set)
+	if got := r.FindByID(set.obj.ID); got.obj.ID != set.obj.ID {
+		t.Fatalf("FindById(%d) failed, got: %+v, expected: %+v", set.obj.ID, got, set)
 	}
 
 	ops := []linux.Sembuf{
@@ -155,14 +157,14 @@ func TestUnregister(t *testing.T) {
 	}
 
 	creds := auth.CredentialsFromContext(ctx)
-	if err := r.RemoveID(set.ID, creds); err != nil {
-		t.Fatalf("RemoveID(%d) failed, err: %v", set.ID, err)
+	if err := r.Remove(set.obj.ID, creds); err != nil {
+		t.Fatalf("Remove(%d) failed, err: %v", set.obj.ID, err)
 	}
 	if !set.dead {
 		t.Fatalf("set is not dead: %+v", set)
 	}
-	if got := r.FindByID(set.ID); got != nil {
-		t.Fatalf("FindById(%d) failed, got: %+v, expected: nil", set.ID, got)
+	if got := r.FindByID(set.obj.ID); got != nil {
+		t.Fatalf("FindById(%d) failed, got: %+v, expected: nil", set.obj.ID, got)
 	}
 	for i, ch := range chs {
 		if !signalled(ch) {
diff --git a/pkg/sentry/kernel/shm/BUILD b/pkg/sentry/kernel/shm/BUILD
index 5b69333fe..4e8deac4c 100644
--- a/pkg/sentry/kernel/shm/BUILD
+++ b/pkg/sentry/kernel/shm/BUILD
@@ -36,6 +36,7 @@ go_library(
         "//pkg/sentry/device",
         "//pkg/sentry/fs",
         "//pkg/sentry/kernel/auth",
+        "//pkg/sentry/kernel/ipc",
         "//pkg/sentry/kernel/time",
         "//pkg/sentry/memmap",
         "//pkg/sentry/pgalloc",
diff --git a/pkg/sentry/kernel/shm/shm.go b/pkg/sentry/kernel/shm/shm.go
index f7ac4c2b2..2abf467d7 100644
--- a/pkg/sentry/kernel/shm/shm.go
+++ b/pkg/sentry/kernel/shm/shm.go
@@ -43,6 +43,7 @@ import (
 	"gvisor.dev/gvisor/pkg/log"
 	"gvisor.dev/gvisor/pkg/sentry/fs"
 	"gvisor.dev/gvisor/pkg/sentry/kernel/auth"
+	"gvisor.dev/gvisor/pkg/sentry/kernel/ipc"
 	ktime "gvisor.dev/gvisor/pkg/sentry/kernel/time"
 	"gvisor.dev/gvisor/pkg/sentry/memmap"
 	"gvisor.dev/gvisor/pkg/sentry/pgalloc"
@@ -51,12 +52,6 @@ import (
 	"gvisor.dev/gvisor/pkg/syserror"
 )
 
-// Key represents a shm segment key. Analogous to a file name.
-type Key int32
-
-// ID represents the opaque handle for a shm segment. Analogous to an fd.
-type ID int32
-
 // Registry tracks all shared memory segments in an IPC namespace. The registry
 // provides the mechanisms for creating and finding segments, and reporting
 // global shm parameters.
@@ -69,50 +64,51 @@ type Registry struct {
 	// mu protects all fields below.
 	mu sync.Mutex `state:"nosave"`
 
-	// shms maps segment ids to segments.
+	// reg defines basic fields and operations needed for all SysV registries.
 	//
-	// shms holds all referenced segments, which are removed on the last
+	// Withing reg, there are two maps, Objects and KeysToIDs.
+	//
+	// reg.objects holds all referenced segments, which are removed on the last
 	// DecRef. Thus, it cannot itself hold a reference on the Shm.
 	//
 	// Since removal only occurs after the last (unlocked) DecRef, there
 	// exists a short window during which a Shm still exists in Shm, but is
 	// unreferenced. Users must use TryIncRef to determine if the Shm is
 	// still valid.
-	shms map[ID]*Shm
-
-	// keysToShms maps segment keys to segments.
 	//
-	// Shms in keysToShms are guaranteed to be referenced, as they are
+	// keysToIDs maps segment keys to IDs.
+	//
+	// Shms in keysToIDs are guaranteed to be referenced, as they are
 	// removed by disassociateKey before the last DecRef.
-	keysToShms map[Key]*Shm
+	reg *ipc.Registry
 
 	// Sum of the sizes of all existing segments rounded up to page size, in
 	// units of page size.
 	totalPages uint64
-
-	// ID assigned to the last created segment. Used to quickly find the next
-	// unused ID.
-	lastIDUsed ID
 }
 
 // NewRegistry creates a new shm registry.
 func NewRegistry(userNS *auth.UserNamespace) *Registry {
 	return &Registry{
-		userNS:     userNS,
-		shms:       make(map[ID]*Shm),
-		keysToShms: make(map[Key]*Shm),
+		userNS: userNS,
+		reg:    ipc.NewRegistry(userNS),
 	}
 }
 
 // FindByID looks up a segment given an ID.
 //
 // FindByID returns a reference on Shm.
-func (r *Registry) FindByID(id ID) *Shm {
+func (r *Registry) FindByID(id ipc.ID) *Shm {
 	r.mu.Lock()
 	defer r.mu.Unlock()
-	s := r.shms[id]
+	mech := r.reg.FindByID(id)
+	if mech == nil {
+		return nil
+	}
+	s := mech.(*Shm)
+
 	// Take a reference on s. If TryIncRef fails, s has reached the last
-	// DecRef, but hasn't quite been removed from r.shms yet.
+	// DecRef, but hasn't quite been removed from r.reg.objects yet.
 	if s != nil && s.TryIncRef() {
 		return s
 	}
@@ -129,9 +125,9 @@ func (r *Registry) dissociateKey(s *Shm) {
 	defer r.mu.Unlock()
 	s.mu.Lock()
 	defer s.mu.Unlock()
-	if s.key != linux.IPC_PRIVATE {
-		delete(r.keysToShms, s.key)
-		s.key = linux.IPC_PRIVATE
+	if s.obj.Key != linux.IPC_PRIVATE {
+		r.reg.DissociateKey(s.obj.Key)
+		s.obj.Key = linux.IPC_PRIVATE
 	}
 }
 
@@ -139,7 +135,7 @@ func (r *Registry) dissociateKey(s *Shm) {
 // analogous to open(2).
 //
 // FindOrCreate returns a reference on Shm.
-func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key Key, size uint64, mode linux.FileMode, private, create, exclusive bool) (*Shm, error) {
+func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key ipc.Key, size uint64, mode linux.FileMode, private, create, exclusive bool) (*Shm, error) {
 	if (create || private) && (size < linux.SHMMIN || size > linux.SHMMAX) {
 		// "A new segment was to be created and size is less than SHMMIN or
 		// greater than SHMMAX." - man shmget(2)
@@ -152,49 +148,29 @@ func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key Key, size ui
 	r.mu.Lock()
 	defer r.mu.Unlock()
 
-	if len(r.shms) >= linux.SHMMNI {
+	if r.reg.ObjectCount() >= linux.SHMMNI {
 		// "All possible shared memory IDs have been taken (SHMMNI) ..."
 		//   - man shmget(2)
 		return nil, syserror.ENOSPC
 	}
 
 	if !private {
-		// Look up an existing segment.
-		if shm := r.keysToShms[key]; shm != nil {
-			shm.mu.Lock()
-			defer shm.mu.Unlock()
-
-			// Check that caller can access the segment.
-			if !shm.checkPermissions(ctx, fs.PermsFromMode(mode)) {
-				// "The user does not have permission to access the shared
-				// memory segment, and does not have the CAP_IPC_OWNER
-				// capability in the user namespace that governs its IPC
-				// namespace." - man shmget(2)
-				return nil, linuxerr.EACCES
-			}
+		shm, err := r.reg.Find(ctx, key, mode, create, exclusive)
+		if err != nil {
+			return nil, err
+		}
 
+		// Validate shm-specific parameters.
+		if shm != nil {
+			shm := shm.(*Shm)
 			if size > shm.size {
 				// "A segment for the given key exists, but size is greater than
 				// the size of that segment." - man shmget(2)
 				return nil, linuxerr.EINVAL
 			}
-
-			if create && exclusive {
-				// "IPC_CREAT and IPC_EXCL were specified in shmflg, but a
-				// shared memory segment already exists for key."
-				//  - man shmget(2)
-				return nil, linuxerr.EEXIST
-			}
-
 			shm.IncRef()
 			return shm, nil
 		}
-
-		if !create {
-			// "No segment exists for the given key, and IPC_CREAT was not
-			// specified." - man shmget(2)
-			return nil, syserror.ENOENT
-		}
 	}
 
 	var sizeAligned uint64
@@ -212,9 +188,7 @@ func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key Key, size ui
 	}
 
 	// Need to create a new segment.
-	creator := fs.FileOwnerFromContext(ctx)
-	perms := fs.FilePermsFromMode(mode)
-	s, err := r.newShm(ctx, pid, key, creator, perms, size)
+	s, err := r.newShmLocked(ctx, pid, key, fs.FileOwnerFromContext(ctx), fs.FilePermsFromMode(mode), size)
 	if err != nil {
 		return nil, err
 	}
@@ -224,10 +198,10 @@ func (r *Registry) FindOrCreate(ctx context.Context, pid int32, key Key, size ui
 	return s, nil
 }
 
-// newShm creates a new segment in the registry.
+// newShmLocked creates a new segment in the registry.
 //
 // Precondition: Caller must hold r.mu.
-func (r *Registry) newShm(ctx context.Context, pid int32, key Key, creator fs.FileOwner, perms fs.FilePermissions, size uint64) (*Shm, error) {
+func (r *Registry) newShmLocked(ctx context.Context, pid int32, key ipc.Key, creator fs.FileOwner, perms fs.FilePermissions, size uint64) (*Shm, error) {
 	mfp := pgalloc.MemoryFileProviderFromContext(ctx)
 	if mfp == nil {
 		panic(fmt.Sprintf("context.Context %T lacks non-nil value for key %T", ctx, pgalloc.CtxMemoryFileProvider))
@@ -242,40 +216,21 @@ func (r *Registry) newShm(ctx context.Context, pid int32, key Key, creator fs.Fi
 	shm := &Shm{
 		mfp:           mfp,
 		registry:      r,
-		creator:       creator,
 		size:          size,
 		effectiveSize: effectiveSize,
+		obj:           ipc.NewObject(r.reg.UserNS, ipc.Key(key), creator, creator, perms),
 		fr:            fr,
-		key:           key,
-		perms:         perms,
-		owner:         creator,
 		creatorPID:    pid,
 		changeTime:    ktime.NowFromContext(ctx),
 	}
 	shm.InitRefs()
 
-	// Find the next available ID.
-	for id := r.lastIDUsed + 1; id != r.lastIDUsed; id++ {
-		// Handle wrap around.
-		if id < 0 {
-			id = 0
-			continue
-		}
-		if r.shms[id] == nil {
-			r.lastIDUsed = id
-
-			shm.ID = id
-			r.shms[id] = shm
-			r.keysToShms[key] = shm
-
-			r.totalPages += effectiveSize / hostarch.PageSize
-
-			return shm, nil
-		}
+	if err := r.reg.Register(shm); err != nil {
+		return nil, err
 	}
+	r.totalPages += effectiveSize / hostarch.PageSize
 
-	log.Warningf("Shm ids exhuasted, they may be leaking")
-	return nil, syserror.ENOSPC
+	return shm, nil
 }
 
 // IPCInfo reports global parameters for sysv shared memory segments on this
@@ -297,7 +252,7 @@ func (r *Registry) ShmInfo() *linux.ShmInfo {
 	defer r.mu.Unlock()
 
 	return &linux.ShmInfo{
-		UsedIDs: int32(r.lastIDUsed),
+		UsedIDs: int32(r.reg.LastIDUsed()),
 		ShmTot:  r.totalPages,
 		ShmRss:  r.totalPages, // We could probably get a better estimate from memory accounting.
 		ShmSwp:  0,            // No reclaim at the moment.
@@ -314,11 +269,11 @@ func (r *Registry) remove(s *Shm) {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 
-	if s.key != linux.IPC_PRIVATE {
+	if s.obj.Key != linux.IPC_PRIVATE {
 		panic(fmt.Sprintf("Attempted to remove %s from the registry whose key is still associated", s.debugLocked()))
 	}
 
-	delete(r.shms, s.ID)
+	r.reg.DissociateID(s.obj.ID)
 	r.totalPages -= s.effectiveSize / hostarch.PageSize
 }
 
@@ -330,13 +285,16 @@ func (r *Registry) Release(ctx context.Context) {
 	// the IPC namespace containing it has no more references.
 	toRelease := make([]*Shm, 0)
 	r.mu.Lock()
-	for _, s := range r.keysToShms {
-		s.mu.Lock()
-		if !s.pendingDestruction {
-			toRelease = append(toRelease, s)
-		}
-		s.mu.Unlock()
-	}
+	r.reg.ForAllObjects(
+		func(o ipc.Mechanism) {
+			s := o.(*Shm)
+			s.mu.Lock()
+			if !s.pendingDestruction {
+				toRelease = append(toRelease, s)
+			}
+			s.mu.Unlock()
+		},
+	)
 	r.mu.Unlock()
 
 	for _, s := range toRelease {
@@ -374,12 +332,6 @@ type Shm struct {
 	// registry points to the shm registry containing this segment. Immutable.
 	registry *Registry
 
-	// ID is the kernel identifier for this segment. Immutable.
-	ID ID
-
-	// creator is the user that created the segment. Immutable.
-	creator fs.FileOwner
-
 	// size is the requested size of the segment at creation, in
 	// bytes. Immutable.
 	size uint64
@@ -397,14 +349,8 @@ type Shm struct {
 	// mu protects all fields below.
 	mu sync.Mutex `state:"nosave"`
 
-	// key is the public identifier for this segment.
-	key Key
-
-	// perms is the access permissions for the segment.
-	perms fs.FilePermissions
+	obj *ipc.Object
 
-	// owner of this segment.
-	owner fs.FileOwner
 	// attachTime is updated on every successful shmat.
 	attachTime ktime.Time
 	// detachTime is updated on every successful shmdt.
@@ -426,17 +372,44 @@ type Shm struct {
 	pendingDestruction bool
 }
 
+// ID returns object's ID.
+func (s *Shm) ID() ipc.ID {
+	return s.obj.ID
+}
+
+// Object implements ipc.Mechanism.Object.
+func (s *Shm) Object() *ipc.Object {
+	return s.obj
+}
+
+// Destroy implements ipc.Mechanism.Destroy. No work is performed on shm.Destroy
+// because a different removal mechanism is used in shm. See Shm.MarkDestroyed.
+func (s *Shm) Destroy() {
+}
+
+// Lock implements ipc.Mechanism.Lock.
+func (s *Shm) Lock() {
+	s.mu.Lock()
+}
+
+// Unlock implements ipc.mechanism.Unlock.
+//
+// +checklocksignore
+func (s *Shm) Unlock() {
+	s.mu.Unlock()
+}
+
 // Precondition: Caller must hold s.mu.
 func (s *Shm) debugLocked() string {
 	return fmt.Sprintf("Shm{id: %d, key: %d, size: %d bytes, refs: %d, destroyed: %v}",
-		s.ID, s.key, s.size, s.ReadRefs(), s.pendingDestruction)
+		s.obj.ID, s.obj.Key, s.size, s.ReadRefs(), s.pendingDestruction)
 }
 
 // MappedName implements memmap.MappingIdentity.MappedName.
 func (s *Shm) MappedName(ctx context.Context) string {
 	s.mu.Lock()
 	defer s.mu.Unlock()
-	return fmt.Sprintf("SYSV%08d", s.key)
+	return fmt.Sprintf("SYSV%08d", s.obj.Key)
 }
 
 // DeviceID implements memmap.MappingIdentity.DeviceID.
@@ -448,7 +421,7 @@ func (s *Shm) DeviceID() uint64 {
 func (s *Shm) InodeID() uint64 {
 	// "shmid gets reported as "inode#" in /proc/pid/maps. proc-ps tools use
 	// this. Changing this will break them." -- Linux, ipc/shm.c:newseg()
-	return uint64(s.ID)
+	return uint64(s.obj.ID)
 }
 
 // DecRef drops a reference on s.
@@ -551,7 +524,8 @@ func (s *Shm) ConfigureAttach(ctx context.Context, addr hostarch.Addr, opts Atta
 		return memmap.MMapOpts{}, syserror.EIDRM
 	}
 
-	if !s.checkPermissions(ctx, fs.PermMask{
+	creds := auth.CredentialsFromContext(ctx)
+	if !s.obj.CheckPermissions(creds, fs.PermMask{
 		Read:    true,
 		Write:   !opts.Readonly,
 		Execute: opts.Execute,
@@ -591,7 +565,8 @@ func (s *Shm) IPCStat(ctx context.Context) (*linux.ShmidDS, error) {
 
 	// "The caller must have read permission on the shared memory segment."
 	//   - man shmctl(2)
-	if !s.checkPermissions(ctx, fs.PermMask{Read: true}) {
+	creds := auth.CredentialsFromContext(ctx)
+	if !s.obj.CheckPermissions(creds, fs.PermMask{Read: true}) {
 		// "IPC_STAT or SHM_STAT is requested and shm_perm.mode does not allow
 		// read access for shmid, and the calling process does not have the
 		// CAP_IPC_OWNER capability in the user namespace that governs its IPC
@@ -603,7 +578,6 @@ func (s *Shm) IPCStat(ctx context.Context) (*linux.ShmidDS, error) {
 	if s.pendingDestruction {
 		mode |= linux.SHM_DEST
 	}
-	creds := auth.CredentialsFromContext(ctx)
 
 	// Use the reference count as a rudimentary count of the number of
 	// attaches. We exclude:
@@ -620,12 +594,12 @@ func (s *Shm) IPCStat(ctx context.Context) (*linux.ShmidDS, error) {
 
 	ds := &linux.ShmidDS{
 		ShmPerm: linux.IPCPerm{
-			Key:  uint32(s.key),
-			UID:  uint32(creds.UserNamespace.MapFromKUID(s.owner.UID)),
-			GID:  uint32(creds.UserNamespace.MapFromKGID(s.owner.GID)),
-			CUID: uint32(creds.UserNamespace.MapFromKUID(s.creator.UID)),
-			CGID: uint32(creds.UserNamespace.MapFromKGID(s.creator.GID)),
-			Mode: mode | uint16(s.perms.LinuxMode()),
+			Key:  uint32(s.obj.Key),
+			UID:  uint32(creds.UserNamespace.MapFromKUID(s.obj.Owner.UID)),
+			GID:  uint32(creds.UserNamespace.MapFromKGID(s.obj.Owner.GID)),
+			CUID: uint32(creds.UserNamespace.MapFromKUID(s.obj.Creator.UID)),
+			CGID: uint32(creds.UserNamespace.MapFromKGID(s.obj.Creator.GID)),
+			Mode: mode | uint16(s.obj.Perms.LinuxMode()),
 			Seq:  0, // IPC sequences not supported.
 		},
 		ShmSegsz:   s.size,
@@ -645,11 +619,11 @@ func (s *Shm) Set(ctx context.Context, ds *linux.ShmidDS) error {
 	s.mu.Lock()
 	defer s.mu.Unlock()
 
-	if !s.checkOwnership(ctx) {
+	creds := auth.CredentialsFromContext(ctx)
+	if !s.obj.CheckOwnership(creds) {
 		return linuxerr.EPERM
 	}
 
-	creds := auth.CredentialsFromContext(ctx)
 	uid := creds.UserNamespace.MapToKUID(auth.UID(ds.ShmPerm.UID))
 	gid := creds.UserNamespace.MapToKGID(auth.GID(ds.ShmPerm.GID))
 	if !uid.Ok() || !gid.Ok() {
@@ -659,10 +633,10 @@ func (s *Shm) Set(ctx context.Context, ds *linux.ShmidDS) error {
 	// User may only modify the lower 9 bits of the mode. All the other bits are
 	// always 0 for the underlying inode.
 	mode := linux.FileMode(ds.ShmPerm.Mode & 0x1ff)
-	s.perms = fs.FilePermsFromMode(mode)
+	s.obj.Perms = fs.FilePermsFromMode(mode)
 
-	s.owner.UID = uid
-	s.owner.GID = gid
+	s.obj.Owner.UID = uid
+	s.obj.Owner.GID = gid
 
 	s.changeTime = ktime.NowFromContext(ctx)
 	return nil
@@ -691,40 +665,3 @@ func (s *Shm) MarkDestroyed(ctx context.Context) {
 	s.DecRef(ctx)
 	return
 }
-
-// checkOwnership verifies whether a segment may be accessed by ctx as an
-// owner. See ipc/util.c:ipcctl_pre_down_nolock() in Linux.
-//
-// Precondition: Caller must hold s.mu.
-func (s *Shm) checkOwnership(ctx context.Context) bool {
-	creds := auth.CredentialsFromContext(ctx)
-	if s.owner.UID == creds.EffectiveKUID || s.creator.UID == creds.EffectiveKUID {
-		return true
-	}
-
-	// Tasks with CAP_SYS_ADMIN may bypass ownership checks. Strangely, Linux
-	// doesn't use CAP_IPC_OWNER for this despite CAP_IPC_OWNER being documented
-	// for use to "override IPC ownership checks".
-	return creds.HasCapabilityIn(linux.CAP_SYS_ADMIN, s.registry.userNS)
-}
-
-// checkPermissions verifies whether a segment is accessible by ctx for access
-// described by req. See ipc/util.c:ipcperms() in Linux.
-//
-// Precondition: Caller must hold s.mu.
-func (s *Shm) checkPermissions(ctx context.Context, req fs.PermMask) bool {
-	creds := auth.CredentialsFromContext(ctx)
-
-	p := s.perms.Other
-	if s.owner.UID == creds.EffectiveKUID {
-		p = s.perms.User
-	} else if creds.InGroup(s.owner.GID) {
-		p = s.perms.Group
-	}
-	if p.SupersetOf(req) {
-		return true
-	}
-
-	// Tasks with CAP_IPC_OWNER may bypass permission checks.
-	return creds.HasCapabilityIn(linux.CAP_IPC_OWNER, s.registry.userNS)
-}