Add task work mechanism.

Like task_work in Linux, this allows us to register callbacks to be executed
before returning to userspace. This is needed for kcov support, which requires
coverage information to be up-to-date whenever we are in user mode. We will
provide coverage data through the kcov interface to enable coverage-directed
fuzzing in syzkaller.

One difference from Linux is that task work cannot queue work before the
transition to userspace that it precedes; queued work will be picked up before
the next transition.

PiperOrigin-RevId: 322889984

pkg/sentry/kernel/BUILD

@@ -132,6 +132,7 @@ go_library(
         "task_stop.go",
         "task_syscall.go",
         "task_usermem.go",
+        "task_work.go",
         "thread_group.go",
         "threads.go",
         "timekeeper.go",

pkg/sentry/kernel/task.go

@@ -68,6 +68,21 @@ type Task struct {
 	// runState is exclusive to the task goroutine.
 	runState taskRunState
 
+	// taskWorkCount represents the current size of the task work queue. It is
+	// used to avoid acquiring taskWorkMu when the queue is empty.
+	//
+	// Must accessed with atomic memory operations.
+	taskWorkCount int32
+
+	// taskWorkMu protects taskWork.
+	taskWorkMu sync.Mutex `state:"nosave"`
+
+	// taskWork is a queue of work to be executed before resuming user execution.
+	// It is similar to the task_work mechanism in Linux.
+	//
+	// taskWork is exclusive to the task goroutine.
+	taskWork []TaskWorker
+
 	// haveSyscallReturn is true if tc.Arch().Return() represents a value
 	// returned by a syscall (or set by ptrace after a syscall).
 	//

pkg/sentry/kernel/task_run.go

@@ -167,7 +167,22 @@ func (app *runApp) execute(t *Task) taskRunState {
 		return (*runInterrupt)(nil)
 	}
 
-	// We're about to switch to the application again. If there's still a
+	// Execute any task work callbacks before returning to user space.
+	if atomic.LoadInt32(&t.taskWorkCount) > 0 {
+		t.taskWorkMu.Lock()
+		queue := t.taskWork
+		t.taskWork = nil
+		atomic.StoreInt32(&t.taskWorkCount, 0)
+		t.taskWorkMu.Unlock()
+
+		// Do not hold taskWorkMu while executing task work, which may register
+		// more work.
+		for _, work := range queue {
+			work.TaskWork(t)
+		}
+	}
+
+	// We're about to switch to the application again. If there's still an
 	// unhandled SyscallRestartErrno that wasn't translated to an EINTR,
 	// restart the syscall that was interrupted. If there's a saved signal
 	// mask, restore it. (Note that restoring the saved signal mask may unblock

pkg/sentry/kernel/task_work.go

@@ -0,0 +1,38 @@
+// Copyright 2020 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 kernel
+
+import "sync/atomic"
+
+// TaskWorker is a deferred task.
+//
+// This must be savable.
+type TaskWorker interface {
+	// TaskWork will be executed prior to returning to user space. Note that
+	// TaskWork may call RegisterWork again, but this will not be executed until
+	// the next return to user space, unlike in Linux. This effectively allows
+	// registration of indefinite user return hooks, but not by default.
+	TaskWork(t *Task)
+}
+
+// RegisterWork can be used to register additional task work that will be
+// performed prior to returning to user space. See TaskWorker.TaskWork for
+// semantics regarding registration.
+func (t *Task) RegisterWork(work TaskWorker) {
+	t.taskWorkMu.Lock()
+	defer t.taskWorkMu.Unlock()
+	atomic.AddInt32(&t.taskWorkCount, 1)
+	t.taskWork = append(t.taskWork, work)
+}