container/op: merge mqueue and dev Ops

There is no reason to mount mqueue anywhere else, and these Ops usually follow each other. This change merges them. This helps decrease IPC overhead and also enables mounting dev readonly.

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

container/container_test.go

@@ -72,8 +72,7 @@ var containerTestCases = []struct {
 
 	{"dev", true, true /* go test output is not a tty */, false, false,
 		new(container.Ops).
-			Dev("/dev").
-			Mqueue("/dev/mqueue"),
+			Dev("/dev", true),
 		[]*vfs.MountInfoEntry{
 			ent("/", "/dev", "rw,nosuid,nodev,relatime", "tmpfs", "devtmpfs", ignore),
 			ent("/null", "/dev/null", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
@@ -86,6 +85,21 @@ var containerTestCases = []struct {
 			ent("/", "/dev/mqueue", "rw,nosuid,nodev,noexec,relatime", "mqueue", "mqueue", "rw"),
 		},
 		1971, 100, nil, 0, seccomp.PresetStrict},
+
+	{"dev no mqueue", true, true /* go test output is not a tty */, false, false,
+		new(container.Ops).
+			Dev("/dev", false),
+		[]*vfs.MountInfoEntry{
+			ent("/", "/dev", "rw,nosuid,nodev,relatime", "tmpfs", "devtmpfs", ignore),
+			ent("/null", "/dev/null", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
+			ent("/zero", "/dev/zero", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
+			ent("/full", "/dev/full", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
+			ent("/random", "/dev/random", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
+			ent("/urandom", "/dev/urandom", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
+			ent("/tty", "/dev/tty", "rw,nosuid", "devtmpfs", "devtmpfs", ignore),
+			ent("/", "/dev/pts", "rw,nosuid,noexec,relatime", "devpts", "devpts", "rw,mode=620,ptmxmode=666"),
+		},
+		1971, 100, nil, 0, seccomp.PresetStrict},
 }
 
 func TestContainer(t *testing.T) {

container/ops.go

@@ -180,27 +180,29 @@ func (p MountProcOp) String() string { return fmt.Sprintf("proc on %q", string(p
 func init() { gob.Register(new(MountDevOp)) }
 
 // Dev appends an [Op] that mounts a subset of host /dev.
-func (f *Ops) Dev(dest string) *Ops {
-	*f = append(*f, MountDevOp(dest))
+func (f *Ops) Dev(dest string, mqueue bool) *Ops {
+	*f = append(*f, &MountDevOp{dest, mqueue})
 	return f
 }
 
-type MountDevOp string
-
-func (d MountDevOp) early(*Params) error { return nil }
-func (d MountDevOp) apply(params *Params) error {
-	v := string(d)
-	if !path.IsAbs(v) {
-		return msg.WrapErr(EBADE, fmt.Sprintf("path %q is not absolute", v))
+type MountDevOp struct {
+	Target string
+	Mqueue bool
 }
-	target := toSysroot(v)
+
+func (d *MountDevOp) early(*Params) error { return nil }
+func (d *MountDevOp) apply(params *Params) error {
+	if !path.IsAbs(d.Target) {
+		return msg.WrapErr(EBADE, fmt.Sprintf("path %q is not absolute", d.Target))
+	}
+	target := toSysroot(d.Target)
 
 	if err := mountTmpfs(SourceTmpfsDevtmpfs, target, MS_NOSUID|MS_NODEV, 0, params.ParentPerm); err != nil {
 		return err
 	}
 
 	for _, name := range []string{"null", "zero", "full", "random", "urandom", "tty"} {
-		targetPath := toSysroot(path.Join(v, name))
+		targetPath := path.Join(target, name)
 		if err := ensureFile(targetPath, 0444, params.ParentPerm); err != nil {
 			return err
 		}
@@ -247,7 +249,7 @@ func (d MountDevOp) apply(params *Params) error {
 	if params.RetainSession {
 		var buf [8]byte
 		if _, _, errno := Syscall(SYS_IOCTL, 1, TIOCGWINSZ, uintptr(unsafe.Pointer(&buf[0]))); errno == 0 {
-			consolePath := toSysroot(path.Join(v, "console"))
+			consolePath := path.Join(target, "console")
 			if err := ensureFile(consolePath, 0444, params.ParentPerm); err != nil {
 				return err
 			}
@@ -264,43 +266,27 @@ func (d MountDevOp) apply(params *Params) error {
 		}
 	}
 
+	if d.Mqueue {
+		mqueueTarget := path.Join(target, "mqueue")
+		if err := os.Mkdir(mqueueTarget, params.ParentPerm); err != nil {
+			return wrapErrSelf(err)
+		}
+		return wrapErrSuffix(Mount(SourceMqueue, mqueueTarget, FstypeMqueue, MS_NOSUID|MS_NOEXEC|MS_NODEV, zeroString),
+			"cannot mount mqueue:")
+	}
+
 	return nil
 }
 
-func (d MountDevOp) Is(op Op) bool  { vd, ok := op.(MountDevOp); return ok && d == vd }
-func (MountDevOp) prefix() string   { return "mounting" }
-func (d MountDevOp) String() string { return fmt.Sprintf("dev on %q", string(d)) }
-
-func init() { gob.Register(new(MountMqueueOp)) }
-
-// Mqueue appends an [Op] that mounts a private instance of mqueue.
-func (f *Ops) Mqueue(dest string) *Ops {
-	*f = append(*f, MountMqueueOp(dest))
-	return f
+func (d *MountDevOp) Is(op Op) bool { vd, ok := op.(*MountDevOp); return ok && *d == *vd }
+func (*MountDevOp) prefix() string  { return "mounting" }
+func (d *MountDevOp) String() string {
+	if d.Mqueue {
+		return fmt.Sprintf("dev on %q with mqueue", d.Target)
 	}
-
-type MountMqueueOp string
-
-func (m MountMqueueOp) early(*Params) error { return nil }
-func (m MountMqueueOp) apply(params *Params) error {
-	v := string(m)
-
-	if !path.IsAbs(v) {
-		return msg.WrapErr(EBADE, fmt.Sprintf("path %q is not absolute", v))
+	return fmt.Sprintf("dev on %q", d.Target)
 }
 
-	target := toSysroot(v)
-	if err := os.MkdirAll(target, params.ParentPerm); err != nil {
-		return wrapErrSelf(err)
-	}
-	return wrapErrSuffix(Mount(SourceMqueue, target, FstypeMqueue, MS_NOSUID|MS_NOEXEC|MS_NODEV, zeroString),
-		fmt.Sprintf("cannot mount mqueue on %q:", v))
-}
-
-func (m MountMqueueOp) Is(op Op) bool  { vm, ok := op.(MountMqueueOp); return ok && m == vm }
-func (MountMqueueOp) prefix() string   { return "mounting" }
-func (m MountMqueueOp) String() string { return fmt.Sprintf("mqueue on %q", string(m)) }
-
 func init() { gob.Register(new(MountTmpfsOp)) }
 
 // Tmpfs appends an [Op] that mounts tmpfs on container path [MountTmpfsOp.Path].

helper/container_test.go

@@ -33,7 +33,7 @@ func TestContainer(t *testing.T) {
 		testHelper(t, func(ctx context.Context, setOutput func(stdoutP, stderrP *io.Writer), stat bool) helper.Helper {
 			return helper.New(ctx, os.Args[0], argsWt, stat, argF, func(z *container.Container) {
 				setOutput(&z.Stdout, &z.Stderr)
-				z.Bind("/", "/", 0).Proc("/proc").Dev("/dev")
+				z.Bind("/", "/", 0).Proc("/proc").Dev("/dev", true)
 			}, nil)
 		})
 	})

internal/app/app_nixos_linux_test.go

@@ -118,7 +118,7 @@ var testCasesNixos = []sealTestCase{
 			Ops: new(container.Ops).
 				Proc("/proc").
 				Tmpfs(hst.Tmp, 4096, 0755).
-				Dev("/dev").Mqueue("/dev/mqueue").
+				Dev("/dev", true).
 				Bind("/bin", "/bin", 0).
 				Bind("/usr/bin", "/usr/bin", 0).
 				Bind("/nix/store", "/nix/store", 0).

internal/app/app_pd_linux_test.go

@@ -46,7 +46,7 @@ var testCasesPd = []sealTestCase{
 				Root("/", "4a450b6596d7bc15bd01780eb9a607ac", container.BindWritable).
 				Proc("/proc").
 				Tmpfs(hst.Tmp, 4096, 0755).
-				Dev("/dev").Mqueue("/dev/mqueue").
+				Dev("/dev", true).
 				Bind("/dev/kvm", "/dev/kvm", container.BindWritable|container.BindDevice|container.BindOptional).
 				Readonly("/var/run/nscd", 0755).
 				Tmpfs("/run/user/1971", 8192, 0755).
@@ -180,7 +180,7 @@ var testCasesPd = []sealTestCase{
 				Root("/", "ebf083d1b175911782d413369b64ce7c", container.BindWritable).
 				Proc("/proc").
 				Tmpfs(hst.Tmp, 4096, 0755).
-				Dev("/dev").Mqueue("/dev/mqueue").
+				Dev("/dev", true).
 				Bind("/dev/dri", "/dev/dri", container.BindWritable|container.BindDevice|container.BindOptional).
 				Bind("/dev/kvm", "/dev/kvm", container.BindWritable|container.BindDevice|container.BindOptional).
 				Readonly("/var/run/nscd", 0755).

internal/app/container_linux.go

@@ -85,7 +85,7 @@ func newContainer(s *hst.ContainerConfig, os sys.State, prefix string, uid, gid
 		Tmpfs(hst.Tmp, 1<<12, 0755)
 
 	if !s.Device {
-		params.Dev("/dev").Mqueue("/dev/mqueue")
+		params.Dev("/dev", true)
 	} else {
 		params.Bind("/dev", "/dev", container.BindWritable|container.BindDevice)
 	}

ldd/exec.go

@@ -28,7 +28,7 @@ func Exec(ctx context.Context, p string) ([]*Entry, error) {
 	stdout, stderr := new(bytes.Buffer), new(bytes.Buffer)
 	z.Stdout = stdout
 	z.Stderr = stderr
-	z.Bind("/", "/", 0).Proc("/proc").Dev("/dev")
+	z.Bind("/", "/", 0).Proc("/proc").Dev("/dev", false)
 
 	if err := z.Start(); err != nil {
 		return nil, err