4 files changed, 51 insertions, 5 deletions
diff --git a/pkg/sentry/kernel/epoll/epoll_state.go b/pkg/sentry/kernel/epoll/epoll_state.go
index a0d35d350..8e9f200d0 100644
--- a/pkg/sentry/kernel/epoll/epoll_state.go
+++ b/pkg/sentry/kernel/epoll/epoll_state.go
@@ -38,11 +38,14 @@ func (e *EventPoll) afterLoad() {
 		}
 	}
 
-	for it := e.waitingList.Front(); it != nil; it = it.Next() {
-		if it.id.File.Readiness(it.mask) != 0 {
-			e.waitingList.Remove(it)
-			e.readyList.PushBack(it)
-			it.curList = &e.readyList
+	for it := e.waitingList.Front(); it != nil; {
+		entry := it
+		it = it.Next()
+
+		if entry.id.File.Readiness(entry.mask) != 0 {
+			e.waitingList.Remove(entry)
+			e.readyList.PushBack(entry)
+			entry.curList = &e.readyList
 			e.Notify(waiter.EventIn)
 		}
 	}
diff --git a/pkg/sentry/kernel/task.go b/pkg/sentry/kernel/task.go
index 2cee2e6ed..c0dbbe890 100644
--- a/pkg/sentry/kernel/task.go
+++ b/pkg/sentry/kernel/task.go
@@ -37,6 +37,7 @@ import (
 	"gvisor.dev/gvisor/pkg/sentry/usage"
 	"gvisor.dev/gvisor/pkg/sentry/vfs"
 	"gvisor.dev/gvisor/pkg/sync"
+	"gvisor.dev/gvisor/pkg/syserror"
 	"gvisor.dev/gvisor/pkg/usermem"
 	"gvisor.dev/gvisor/pkg/waiter"
 )
@@ -554,6 +555,13 @@ type Task struct {
 	//
 	// startTime is protected by mu.
 	startTime ktime.Time
+
+	// oomScoreAdj is the task's OOM score adjustment. This is currently not
+	// used but is maintained for consistency.
+	// TODO(gvisor.dev/issue/1967)
+	//
+	// oomScoreAdj is protected by mu, and is owned by the task goroutine.
+	oomScoreAdj int32
 }
 
 func (t *Task) savePtraceTracer() *Task {
@@ -847,3 +855,28 @@ func (t *Task) AbstractSockets() *AbstractSocketNamespace {
 func (t *Task) ContainerID() string {
 	return t.containerID
 }
+
+// OOMScoreAdj gets the task's OOM score adjustment.
+func (t *Task) OOMScoreAdj() (int32, error) {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	if t.ExitState() == TaskExitDead {
+		return 0, syserror.ESRCH
+	}
+	return t.oomScoreAdj, nil
+}
+
+// SetOOMScoreAdj sets the task's OOM score adjustment. The value should be
+// between -1000 and 1000 inclusive.
+func (t *Task) SetOOMScoreAdj(adj int32) error {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	if t.ExitState() == TaskExitDead {
+		return syserror.ESRCH
+	}
+	if adj > 1000 || adj < -1000 {
+		return syserror.EINVAL
+	}
+	t.oomScoreAdj = adj
+	return nil
+}
diff --git a/pkg/sentry/kernel/task_clone.go b/pkg/sentry/kernel/task_clone.go
index 78866f280..dda502bb8 100644
--- a/pkg/sentry/kernel/task_clone.go
+++ b/pkg/sentry/kernel/task_clone.go
@@ -264,6 +264,11 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
 		rseqSignature = t.rseqSignature
 	}
 
+	adj, err := t.OOMScoreAdj()
+	if err != nil {
+		return 0, nil, err
+	}
+
 	cfg := &TaskConfig{
 		Kernel:                  t.k,
 		ThreadGroup:             tg,
@@ -282,6 +287,7 @@ func (t *Task) Clone(opts *CloneOptions) (ThreadID, *SyscallControl, error) {
 		RSeqAddr:                rseqAddr,
 		RSeqSignature:           rseqSignature,
 		ContainerID:             t.ContainerID(),
+		OOMScoreAdj:             adj,
 	}
 	if opts.NewThreadGroup {
 		cfg.Parent = t
diff --git a/pkg/sentry/kernel/task_start.go b/pkg/sentry/kernel/task_start.go
index a5035bb7f..2bbf48bb8 100644
--- a/pkg/sentry/kernel/task_start.go
+++ b/pkg/sentry/kernel/task_start.go
@@ -93,6 +93,9 @@ type TaskConfig struct {
 
 	// ContainerID is the container the new task belongs to.
 	ContainerID string
+
+	// oomScoreAdj is the task's OOM score adjustment.
+	OOMScoreAdj int32
 }
 
 // NewTask creates a new task defined by cfg.
@@ -143,6 +146,7 @@ func (ts *TaskSet) newTask(cfg *TaskConfig) (*Task, error) {
 		rseqSignature:      cfg.RSeqSignature,
 		futexWaiter:        futex.NewWaiter(),
 		containerID:        cfg.ContainerID,
+		oomScoreAdj:        cfg.OOMScoreAdj,
 	}
 	t.creds.Store(cfg.Credentials)
 	t.endStopCond.L = &t.tg.signalHandlers.mu