diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go
index 3e494178d..b9ede9006 100644
--- a/pkg/sentry/kernel/task_clone.go
+++ b/pkg/sentry/kernel/task_clone.go
@@ -147,7 +147,7 @@ func (t *Task) Clone(args *linux.CloneArgs) (ThreadID, *SyscallControl, error) {
 		return 0, nil, err
 	}
 	cu.Add(func() {
-		image.release()
+		image.release(t)
 	})
 	// clone() returns 0 in the child.
 	image.Arch.SetReturn(0)
diff --git a/pkg/sentry/kernel/task_exec.go b/pkg/sentry/kernel/task_exec.go
index 6960c6e6d..460058331 100644
--- a/pkg/sentry/kernel/task_exec.go
+++ b/pkg/sentry/kernel/task_exec.go
@@ -94,7 +94,7 @@ func (*execStop) Killable() bool { return true }
 // goroutine.
 func (t *Task) Execve(newImage *TaskImage, argv, env []string, executable *vfs.FileDescription, pathname string) (*SyscallControl, error) {
 	cu := cleanup.Make(func() {
-		newImage.release()
+		newImage.release(t)
 	})
 	defer cu.Clean()
 	// We can't clearly hold kernel package locks while stat'ing executable.
@@ -154,7 +154,7 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState {
 	t.tg.execing = nil
 	if t.killed() {
 		t.tg.pidns.owner.mu.Unlock()
-		r.image.release()
+		r.image.release(t)
 		return (*runInterrupt)(nil)
 	}
 	// We are the thread group leader now. Save our old thread ID for
@@ -248,7 +248,7 @@ func (r *runSyscallAfterExecStop) execute(t *Task) taskRunState {
 	// Don't hold t.mu while calling t.image.release(), that may
 	// attempt to acquire TaskImage.MemoryManager.mappingMu, a lock order
 	// violation.
-	oldImage.release()
+	oldImage.release(t)
 
 	t.unstopVforkParent()
 	t.p.FullStateChanged()
diff --git a/pkg/sentry/kernel/task_image.go b/pkg/sentry/kernel/task_image.go
index 1e75e9f05..d3c5aee88 100644
--- a/pkg/sentry/kernel/task_image.go
+++ b/pkg/sentry/kernel/task_image.go
@@ -54,11 +54,11 @@ type TaskImage struct {
 
 // release releases all resources held by the TaskImage. release is called by
 // the task when it execs into a new TaskImage.
-func (image *TaskImage) release() {
+func (image *TaskImage) release(ctx context.Context) {
 	// Nil out pointers so that if the task is saved after release, it doesn't
 	// follow the pointers to possibly now-invalid objects.
 	if image.MemoryManager != nil {
-		image.MemoryManager.DecUsers(context.Background())
+		image.MemoryManager.DecUsers(ctx)
 		image.MemoryManager = nil
 	}
 	image.fu = nil
diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go
index ae8f70ad9..5745afc31 100644
--- a/pkg/sentry/kernel/task_start.go
+++ b/pkg/sentry/kernel/task_start.go
@@ -113,7 +113,7 @@ type TaskConfig struct {
 func (ts *TaskSet) NewTask(ctx context.Context, cfg *TaskConfig) (*Task, error) {
 	var err error
 	cleanup := func() {
-		cfg.TaskImage.release()
+		cfg.TaskImage.release(ctx)
 		cfg.FSContext.DecRef(ctx)
 		cfg.FDTable.DecRef(ctx)
 		cfg.IPCNamespace.DecRef(ctx)