container: sched policy string representation

This also uses priority obtained via sched_get_priority_min, and improves bounds checking.

Signed-off-by: Ophestra <cat@gensokyo.uk>

4 files changed, 235 insertions, 7 deletions

diff --git a/container/container.go b/container/container.go
index 8b5ec01d..e7fd8cee 100644
--- a/container/container.go
+++ b/container/container.go
@@ -40,7 +40,7 @@ type (
 		AllowOrphan bool
 		// Scheduling policy to set via sched_setscheduler(2). The zero value
 		// skips this call. Supported policies are [SCHED_BATCH], [SCHED_IDLE].
-		SchedPolicy int
+		SchedPolicy SchedPolicy
 		// Cgroup fd, nil to disable.
 		Cgroup *int
 		// ExtraFiles passed through to initial process in the container, with
@@ -373,12 +373,23 @@ func (p *Container) Start() error {
 
 			// sched_setscheduler: thread-directed but acts on all processes
 			// created from the calling thread
-			if p.SchedPolicy > 0 {
+			if p.SchedPolicy > 0 && p.SchedPolicy <= _SCHED_LAST {
+				var param schedParam
+				if priority, err := p.SchedPolicy.GetPriorityMin(); err != nil {
+					return &StartError{
+						Fatal: true,
+						Step:  "get minimum priority",
+						Err:   err,
+					}
+				} else {
+					param.priority = priority
+				}
+
 				p.msg.Verbosef("setting scheduling policy %d", p.SchedPolicy)
 				if err := schedSetscheduler(
 					0, // calling thread
 					p.SchedPolicy,
-					&schedParam{0},
+					&param,
 				); err != nil {
 					return &StartError{
 						Fatal: true,
diff --git a/container/syscall.go b/container/syscall.go
index 5f392c35..4af60366 100644
--- a/container/syscall.go
+++ b/container/syscall.go
@@ -1,6 +1,9 @@
 package container
 
 import (
+	"encoding"
+	"strconv"
+	"sync"
 	. "syscall"
 	"unsafe"
 
@@ -43,18 +46,132 @@ func Isatty(fd int) bool {
 	return r == 0
 }
 
+// SchedPolicy denotes a scheduling policy defined in include/uapi/linux/sched.h.
+type SchedPolicy int
+
 // include/uapi/linux/sched.h
 const (
-	SCHED_NORMAL = iota
+	SCHED_NORMAL SchedPolicy = iota
 	SCHED_FIFO
 	SCHED_RR
 	SCHED_BATCH
-	_ // SCHED_ISO: reserved but not implemented yet
+	_SCHED_ISO // SCHED_ISO: reserved but not implemented yet
 	SCHED_IDLE
 	SCHED_DEADLINE
 	SCHED_EXT
+
+	_SCHED_LAST SchedPolicy = iota - 1
 )
 
+var _ encoding.TextMarshaler = _SCHED_LAST
+var _ encoding.TextUnmarshaler = new(SchedPolicy)
+
+// String returns a unique representation of policy, also used in encoding.
+func (policy SchedPolicy) String() string {
+	switch policy {
+	case SCHED_NORMAL:
+		return ""
+	case SCHED_FIFO:
+		return "fifo"
+	case SCHED_RR:
+		return "rr"
+	case SCHED_BATCH:
+		return "batch"
+	case SCHED_IDLE:
+		return "idle"
+	case SCHED_DEADLINE:
+		return "deadline"
+	case SCHED_EXT:
+		return "ext"
+
+	default:
+		return "invalid policy " + strconv.Itoa(int(policy))
+	}
+}
+
+// MarshalText performs bounds checking and returns the result of String.
+func (policy SchedPolicy) MarshalText() ([]byte, error) {
+	if policy == _SCHED_ISO || policy < 0 || policy > _SCHED_LAST {
+		return nil, EINVAL
+	}
+	return []byte(policy.String()), nil
+}
+
+// InvalidSchedPolicyError is an invalid string representation of a [SchedPolicy].
+type InvalidSchedPolicyError string
+
+func (InvalidSchedPolicyError) Unwrap() error { return EINVAL }
+func (e InvalidSchedPolicyError) Error() string {
+	return "invalid scheduling policy " + strconv.Quote(string(e))
+}
+
+// UnmarshalText is the inverse of MarshalText.
+func (policy *SchedPolicy) UnmarshalText(text []byte) error {
+	switch string(text) {
+	case "fifo":
+		*policy = SCHED_FIFO
+	case "rr":
+		*policy = SCHED_RR
+	case "batch":
+		*policy = SCHED_BATCH
+	case "idle":
+		*policy = SCHED_IDLE
+	case "deadline":
+		*policy = SCHED_DEADLINE
+	case "ext":
+		*policy = SCHED_EXT
+
+	case "":
+		*policy = 0
+		return nil
+	default:
+		return InvalidSchedPolicyError(text)
+	}
+	return nil
+}
+
+// for sched_get_priority_max and sched_get_priority_min
+var (
+	schedPriority     [_SCHED_LAST + 1][2]std.Int
+	schedPriorityErr  [_SCHED_LAST + 1][2]error
+	schedPriorityOnce [_SCHED_LAST + 1][2]sync.Once
+)
+
+// GetPriorityMax returns the maximum priority value that can be used with the
+// scheduling algorithm identified by policy.
+func (policy SchedPolicy) GetPriorityMax() (std.Int, error) {
+	schedPriorityOnce[policy][0].Do(func() {
+		priority, _, errno := Syscall(
+			SYS_SCHED_GET_PRIORITY_MAX,
+			uintptr(policy),
+			0, 0,
+		)
+		schedPriority[policy][0] = std.Int(priority)
+		if schedPriority[policy][0] < 0 {
+			schedPriorityErr[policy][0] = errno
+		}
+	})
+	return schedPriority[policy][0], schedPriorityErr[policy][0]
+}
+
+// GetPriorityMin returns the minimum priority value that can be used with the
+// scheduling algorithm identified by policy.
+func (policy SchedPolicy) GetPriorityMin() (std.Int, error) {
+	schedPriorityOnce[policy][1].Do(func() {
+		priority, _, errno := Syscall(
+			SYS_SCHED_GET_PRIORITY_MIN,
+			uintptr(policy),
+			0, 0,
+		)
+		schedPriority[policy][1] = std.Int(priority)
+		if schedPriority[policy][1] < 0 {
+			schedPriorityErr[policy][1] = errno
+		}
+	})
+	return schedPriority[policy][1], schedPriorityErr[policy][1]
+
+}
+
 // schedParam is equivalent to struct sched_param from include/linux/sched.h.
 type schedParam struct {
 	// sched_priority
@@ -74,7 +191,7 @@ type schedParam struct {
 // this if you do not have something similar in place!
 //
 // [very subtle to use correctly]: https://www.openwall.com/lists/musl/2016/03/01/4
-func schedSetscheduler(tid, policy int, param *schedParam) error {
+func schedSetscheduler(tid int, policy SchedPolicy, param *schedParam) error {
 	if r, _, errno := Syscall(
 		SYS_SCHED_SETSCHEDULER,
 		uintptr(tid),
diff --git a/container/syscall_test.go b/container/syscall_test.go
new file mode 100644
index 00000000..337d6450
--- /dev/null
+++ b/container/syscall_test.go
@@ -0,0 +1,100 @@
+package container_test
+
+import (
+	"encoding/json"
+	"errors"
+	"math"
+	"reflect"
+	"syscall"
+	"testing"
+
+	"hakurei.app/container"
+	"hakurei.app/container/std"
+)
+
+func TestSchedPolicyJSON(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		policy    container.SchedPolicy
+		want      string
+		encodeErr error
+		decodeErr error
+	}{
+		{container.SCHED_NORMAL, `""`, nil, nil},
+		{container.SCHED_FIFO, `"fifo"`, nil, nil},
+		{container.SCHED_RR, `"rr"`, nil, nil},
+		{container.SCHED_BATCH, `"batch"`, nil, nil},
+		{4, `"invalid policy 4"`, syscall.EINVAL, container.InvalidSchedPolicyError("invalid policy 4")},
+		{container.SCHED_IDLE, `"idle"`, nil, nil},
+		{container.SCHED_DEADLINE, `"deadline"`, nil, nil},
+		{container.SCHED_EXT, `"ext"`, nil, nil},
+		{math.MaxInt, `"iso"`, syscall.EINVAL, container.InvalidSchedPolicyError("iso")},
+	}
+	for _, tc := range testCases {
+		name := tc.policy.String()
+		if tc.policy == container.SCHED_NORMAL {
+			name = "normal"
+		}
+
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			got, err := json.Marshal(tc.policy)
+			if !errors.Is(err, tc.encodeErr) {
+				t.Fatalf("Marshal: error = %v, want %v", err, tc.encodeErr)
+			}
+			if err == nil && string(got) != tc.want {
+				t.Fatalf("Marshal: %s, want %s", string(got), tc.want)
+			}
+
+			var v container.SchedPolicy
+			if err = json.Unmarshal([]byte(tc.want), &v); !reflect.DeepEqual(err, tc.decodeErr) {
+				t.Fatalf("Unmarshal: error = %v, want %v", err, tc.decodeErr)
+			}
+			if err == nil && v != tc.policy {
+				t.Fatalf("Unmarshal: %d, want %d", v, tc.policy)
+			}
+		})
+	}
+}
+
+func TestSchedPolicyMinMax(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		policy   container.SchedPolicy
+		min, max std.Int
+		err      error
+	}{
+		{container.SCHED_NORMAL, 0, 0, nil},
+		{container.SCHED_FIFO, 1, 99, nil},
+		{container.SCHED_RR, 1, 99, nil},
+		{container.SCHED_BATCH, 0, 0, nil},
+		{4, -1, -1, syscall.EINVAL},
+		{container.SCHED_IDLE, 0, 0, nil},
+		{container.SCHED_DEADLINE, 0, 0, nil},
+		{container.SCHED_EXT, 0, 0, nil},
+	}
+	for _, tc := range testCases {
+		name := tc.policy.String()
+		if tc.policy == container.SCHED_NORMAL {
+			name = "normal"
+		}
+
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			if priority, err := tc.policy.GetPriorityMax(); !reflect.DeepEqual(err, tc.err) {
+				t.Fatalf("GetPriorityMax: error = %v, want %v", err, tc.err)
+			} else if priority != tc.max {
+				t.Fatalf("GetPriorityMax: %d, want %d", priority, tc.max)
+			}
+			if priority, err := tc.policy.GetPriorityMin(); !reflect.DeepEqual(err, tc.err) {
+				t.Fatalf("GetPriorityMin: error = %v, want %v", err, tc.err)
+			} else if priority != tc.min {
+				t.Fatalf("GetPriorityMin: %d, want %d", priority, tc.min)
+			}
+		})
+	}
+}
diff --git a/internal/pkg/exec.go b/internal/pkg/exec.go
index eb72ccbd..c7248f1a 100644
--- a/internal/pkg/exec.go
+++ b/internal/pkg/exec.go
@@ -40,7 +40,7 @@ type ExecPath struct {
 }
 
 // SchedPolicy is the [container] scheduling policy.
-var SchedPolicy int
+var SchedPolicy container.SchedPolicy
 
 // PromoteLayers returns artifacts with identical-by-content layers promoted to
 // the highest priority instance, as if mounted via [ExecPath].