internal/pkg: correctly generate cure expects

This needs to dereference the identifier symlink.

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

container/binfmt.go

@@ -0,0 +1,46 @@
+package container
+
+import (
+	"strings"
+	"unsafe"
+
+	"hakurei.app/check"
+)
+
+// escapeBinfmt escapes magic/mask sequences in a [BinfmtEntry].
+func escapeBinfmt(buf *strings.Builder, s string) string {
+	const lowerhex = "0123456789abcdef"
+
+	buf.Reset()
+	for _, c := range unsafe.Slice(unsafe.StringData(s), len(s)) {
+		switch c {
+		case 0, '\\', ':':
+			buf.WriteString(`\x`)
+			buf.WriteByte(lowerhex[c>>4])
+			buf.WriteByte(lowerhex[c&0xf])
+
+		default:
+			buf.WriteByte(c)
+		}
+	}
+	return buf.String()
+}
+
+// BinfmtEntry is an entry to be registered by the init process.
+type BinfmtEntry struct {
+	// The offset of the magic/mask in the file, counted in bytes.
+	Offset byte
+	// The byte sequence binfmt_misc is matching for.
+	Magic string
+	// An (optional, defaults to all 0xff) mask.
+	Mask string
+	// The program that should be invoked with the binary as first argument.
+	Interpreter *check.Absolute
+}
+
+// Valid returns whether e can be registered into the kernel.
+func (e *BinfmtEntry) Valid() bool {
+	return e != nil &&
+		int(e.Offset)+max(len(e.Magic), len(e.Mask)) < 128 &&
+		e.Interpreter != nil && len(e.Interpreter.String()) < 128
+}

container/binfmt_test.go

@@ -0,0 +1,62 @@
+package container
+
+import (
+	"strings"
+	"testing"
+
+	"hakurei.app/fhs"
+)
+
+func TestEscapeBinfmt(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name  string
+		magic string
+		want  string
+	}{
+		{"packed DOS applications", "\x0eDEX", "\x0eDEX"},
+
+		{"riscv64 magic",
+			"\x7fELF\x02\x01\x01\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\xf3\x00",
+			"\x7fELF\x02\x01\x01\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\x02\\x00\xf3\\x00"},
+		{"riscv64 mask",
+			"\xff\xff\xff\xff\xff\xff\xff\x00\xff\xff\xff\xff\xff\xff\xff\xff\xfe\xff\xff\xff",
+			"\xff\xff\xff\xff\xff\xff\xff\\x00\xff\xff\xff\xff\xff\xff\xff\xff\xfe\xff\xff\xff"},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			got := escapeBinfmt(new(strings.Builder), tc.magic)
+			if got != tc.want {
+				t.Errorf("escapeBinfmt: %q, want %q", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestBinfmtEntry(t *testing.T) {
+	t.Parallel()
+
+	testCases := []struct {
+		name  string
+		e     BinfmtEntry
+		valid bool
+	}{
+		{"zero", BinfmtEntry{}, false},
+		{"large offset", BinfmtEntry{Offset: 128}, false},
+		{"long magic", BinfmtEntry{Magic: strings.Repeat("\x00", 128)}, false},
+		{"long mask", BinfmtEntry{Mask: strings.Repeat("\x00", 128)}, false},
+		{"valid", BinfmtEntry{Interpreter: fhs.AbsRoot}, true},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			if tc.e.Valid() != tc.valid {
+				t.Errorf("Valid: %v", !tc.valid)
+			}
+		})
+	}
+}

container/capability.go

@@ -18,6 +18,7 @@ const (
 	CAP_SETPCAP      = 0x8
 	CAP_NET_ADMIN    = 0xc
 	CAP_DAC_OVERRIDE = 0x1
+	CAP_SETFCAP      = 0x1f
 )
 
 type (

container/container.go

@@ -91,12 +91,20 @@ type (
 		// Time to wait for processes lingering after the initial process terminates.
 		AdoptWaitDelay time.Duration
 
+		// Map uid/gid 0 in the init process. Requires [FstypeProc] attached to
+		// [fhs.Proc] in the container filesystem.
+		InitAsRoot bool
 		// Mapped Uid in user namespace.
 		Uid int
 		// Mapped Gid in user namespace.
 		Gid int
 		// Hostname value in UTS namespace.
 		Hostname string
+		// Register binfmt_misc entries.
+		Binfmt []BinfmtEntry
+		// Alternative pathname to attach binfmt_misc filesystem. The zero value
+		// requires [FstypeProc] to be made available at [fhs.Proc].
+		BinfmtPath *check.Absolute
 		// Sequential container setup ops.
 		*Ops
 
@@ -216,6 +224,9 @@ func (p *Container) Start() error {
 	if p.cmd.Process != nil {
 		return errors.New("container: already started")
 	}
+	if !p.InitAsRoot && len(p.Binfmt) > 0 {
+		return errors.New("container: init as root required, but not enabled")
+	}
 
 	if err := ensureCloseOnExec(); err != nil {
 		return err
@@ -286,6 +297,18 @@ func (p *Container) Start() error {
 	if !p.HostNet {
 		p.cmd.SysProcAttr.Cloneflags |= CLONE_NEWNET
 	}
+	if p.InitAsRoot {
+		p.cmd.SysProcAttr.AmbientCaps = append(p.cmd.SysProcAttr.AmbientCaps,
+			// mappings during init as root
+			CAP_SETFCAP,
+		)
+
+		if !p.SeccompDisable &&
+			len(p.SeccompRules) == 0 &&
+			p.SeccompPresets&std.PresetDenyNS != 0 {
+			return errors.New("container: as root requires late namespace creation")
+		}
+	}
 
 	// place setup pipe before user supplied extra files, this is later restored by init
 	if r, w, err := os.Pipe(); err != nil {

container/container_test.go

@@ -16,7 +16,8 @@ import (
 	"strings"
 	"syscall"
 	"testing"
-	_ "unsafe" // for go:linkname
+	"time"
+	"unsafe"
 
 	"hakurei.app/check"
 	"hakurei.app/command"
@@ -408,39 +409,11 @@ var containerTestCases = []struct {
 func TestContainer(t *testing.T) {
 	t.Parallel()
 
-	t.Run("cancel", testContainerCancel(nil, func(t *testing.T, c *container.Container) {
+	var suffix string
-		wantErr := context.Canceled
+runTests:
-		wantExitCode := 0
-		if err := c.Wait(); !reflect.DeepEqual(err, wantErr) {
-			if m, ok := container.InternalMessageFromError(err); ok {
-				t.Error(m)
-			}
-			t.Errorf("Wait: error = %#v, want %#v", err, wantErr)
-		}
-		if ps := c.ProcessState(); ps == nil {
-			t.Errorf("ProcessState unexpectedly returned nil")
-		} else if code := ps.ExitCode(); code != wantExitCode {
-			t.Errorf("ExitCode: %d, want %d", code, wantExitCode)
-		}
-	}))
-
-	t.Run("forward", testContainerCancel(func(c *container.Container) {
-		c.ForwardCancel = true
-	}, func(t *testing.T, c *container.Container) {
-		var exitError *exec.ExitError
-		if err := c.Wait(); !errors.As(err, &exitError) {
-			if m, ok := container.InternalMessageFromError(err); ok {
-				t.Error(m)
-			}
-			t.Errorf("Wait: error = %v", err)
-		}
-		if code := exitError.ExitCode(); code != blockExitCodeInterrupt {
-			t.Errorf("ExitCode: %d, want %d", code, blockExitCodeInterrupt)
-		}
-	}))
-
 	for i, tc := range containerTestCases {
-		t.Run(tc.name, func(t *testing.T) {
+		_suffix := suffix
+		t.Run(tc.name+_suffix, func(t *testing.T) {
 			t.Parallel()
 
 			wantOps, wantOpsCtx := tc.ops(t)
@@ -464,6 +437,8 @@ func TestContainer(t *testing.T) {
 			c.SeccompDisable = !tc.filter
 			c.RetainSession = tc.session
 			c.HostNet = tc.net
+			c.InitAsRoot = _suffix != ""
+			c.Env = append(c.Env, "HAKUREI_TEST_SUFFIX="+_suffix)
 			if info.CanDegrade {
 				if _, err := landlock.GetABI(); err != nil {
 					if !errors.Is(err, syscall.ENOSYS) {
@@ -473,6 +448,9 @@ func TestContainer(t *testing.T) {
 					t.Log("Landlock LSM is unavailable, enabling HostAbstract")
 				}
 			}
+			if c.InitAsRoot {
+				c.SeccompPresets &= ^std.PresetDenyNS
+			}
 
 			c.
 				Readonly(check.MustAbs(pathReadonly), 0755).
@@ -541,6 +519,11 @@ func TestContainer(t *testing.T) {
 			}
 		})
 	}
+
+	if suffix == "" {
+		suffix = " as root"
+		goto runTests
+	}
 }
 
 func ent(root, target, vfsOptstr, fsType, source, fsOptstr string) *vfs.MountInfoEntry {
@@ -563,49 +546,118 @@ func hostnameFromTestCase(name string) string {
 }
 
 func testContainerCancel(
+	t *testing.T,
 	containerExtra func(c *container.Container),
-	waitCheck func(t *testing.T, c *container.Container),
+	waitCheck func(ps *os.ProcessState, waitErr error),
-) func(t *testing.T) {
+) {
-	return func(t *testing.T) {
+	ctx, cancel := context.WithCancel(t.Context())
-		t.Parallel()
-		ctx, cancel := context.WithCancel(t.Context())
 
-		c := helperNewContainer(ctx, "block")
+	c := helperNewContainer(ctx, "block")
-		c.Stdout, c.Stderr = os.Stdout, os.Stderr
+	c.Stdout, c.Stderr = os.Stdout, os.Stderr
-		if containerExtra != nil {
+	if containerExtra != nil {
-			containerExtra(c)
+		containerExtra(c)
-		}
-
-		ready := make(chan struct{})
-		if r, w, err := os.Pipe(); err != nil {
-			t.Fatalf("cannot pipe: %v", err)
-		} else {
-			c.ExtraFiles = append(c.ExtraFiles, w)
-			go func() {
-				defer close(ready)
-				if _, err = r.Read(make([]byte, 1)); err != nil {
-					panic(err.Error())
-				}
-			}()
-		}
-
-		if err := c.Start(); err != nil {
-			if m, ok := container.InternalMessageFromError(err); ok {
-				t.Fatal(m)
-			} else {
-				t.Fatalf("cannot start container: %v", err)
-			}
-		} else if err = c.Serve(); err != nil {
-			if m, ok := container.InternalMessageFromError(err); ok {
-				t.Error(m)
-			} else {
-				t.Errorf("cannot serve setup params: %v", err)
-			}
-		}
-		<-ready
-		cancel()
-		waitCheck(t, c)
 	}
+
+	ready := make(chan struct{})
+	var waitErr error
+	r, w, err := os.Pipe()
+	if err != nil {
+		t.Fatalf("cannot pipe: %v", err)
+	}
+
+	c.ExtraFiles = append(c.ExtraFiles, w)
+	go func() {
+		defer close(ready)
+		if _, _err := r.Read(make([]byte, 1)); _err != nil {
+			panic(_err)
+		}
+	}()
+
+	if err = c.Start(); err != nil {
+		if m, ok := container.InternalMessageFromError(err); ok {
+			t.Fatal(m)
+		} else {
+			t.Fatalf("cannot start container: %v", err)
+		}
+	}
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		waitErr = c.Wait()
+		_ = r.SetReadDeadline(time.Now())
+	}()
+
+	if err = c.Serve(); err != nil {
+		if m, ok := container.InternalMessageFromError(err); ok {
+			t.Error(m)
+		} else {
+			t.Errorf("cannot serve setup params: %v", err)
+		}
+	}
+	<-ready
+	cancel()
+	<-done
+	waitCheck(c.ProcessState(), waitErr)
+}
+
+func TestForward(t *testing.T) {
+	t.Parallel()
+
+	f := func(ps *os.ProcessState, waitErr error) {
+		var exitError *exec.ExitError
+		if !errors.As(waitErr, &exitError) {
+			if m, ok := container.InternalMessageFromError(waitErr); ok {
+				t.Error(m)
+			}
+			t.Errorf("Wait: error = %v", waitErr)
+		}
+		if code := exitError.ExitCode(); code != blockExitCodeInterrupt {
+			t.Errorf("ExitCode: %d, want %d", code, blockExitCodeInterrupt)
+		}
+	}
+	t.Run("direct", func(t *testing.T) {
+		t.Parallel()
+		testContainerCancel(t, func(c *container.Container) {
+			c.ForwardCancel = true
+		}, f)
+	})
+	t.Run("as root", func(t *testing.T) {
+		testContainerCancel(t, func(c *container.Container) {
+			c.ForwardCancel = true
+			c.InitAsRoot = true
+			c.Proc(fhs.AbsProc)
+		}, f)
+	})
+}
+
+func TestCancel(t *testing.T) {
+	t.Parallel()
+
+	f := func(ps *os.ProcessState, waitErr error) {
+		wantErr := context.Canceled
+		if !reflect.DeepEqual(waitErr, wantErr) {
+			if m, ok := container.InternalMessageFromError(waitErr); ok {
+				t.Error(m)
+			}
+			t.Errorf("Wait: error = %#v, want %#v", waitErr, wantErr)
+		}
+		if ps == nil {
+			t.Errorf("ProcessState unexpectedly returned nil")
+		} else if code := ps.ExitCode(); code != 0 {
+			t.Errorf("ExitCode: %d, want %d", code, 0)
+		}
+	}
+	t.Run("direct", func(t *testing.T) {
+		t.Parallel()
+		testContainerCancel(t, nil, f)
+	})
+	t.Run("as root", func(t *testing.T) {
+		testContainerCancel(t, func(c *container.Container) {
+			c.InitAsRoot = true
+			c.Proc(fhs.AbsProc)
+		}, f)
+	})
 }
 
 func TestContainerString(t *testing.T) {
@@ -641,6 +693,8 @@ func init() {
 		})
 
 		c.Command("container", command.UsageInternal, func(args []string) error {
+			asRoot := os.Getenv("HAKUREI_TEST_SUFFIX") == " as root"
+
 			if len(args) != 1 {
 				return syscall.EINVAL
 			}
@@ -658,6 +712,66 @@ func init() {
 				return fmt.Errorf("gid: %d, want %d", gid, tc.gid)
 			}
 
+			// no attack surface increase during as root due to no_new_privs
+			var wantBounding uintptr = 1
+			asRootNot := " not"
+			if !asRoot {
+				wantBounding = 0
+				asRootNot = ""
+			}
+
+			const (
+				PR_CAP_AMBIENT        = 0x2f
+				PR_CAP_AMBIENT_IS_SET = 0x1
+			)
+			for i := range container.LastCap(nil) + 1 {
+				r, _, errno := syscall.Syscall(
+					syscall.SYS_PRCTL,
+					PR_CAP_AMBIENT,
+					PR_CAP_AMBIENT_IS_SET,
+					i,
+				)
+				if errno != 0 {
+					return os.NewSyscallError("prctl", errno)
+				}
+				if r != 0 {
+					return fmt.Errorf("capability %d in ambient set", i)
+				}
+
+				r, _, errno = syscall.Syscall(
+					syscall.SYS_PRCTL,
+					syscall.PR_CAPBSET_READ,
+					i,
+					0,
+				)
+				if errno != 0 {
+					return os.NewSyscallError("prctl", errno)
+				}
+				if r != wantBounding {
+					return fmt.Errorf("capability %d%s in bounding set", i, asRootNot)
+				}
+			}
+
+			const _LINUX_CAPABILITY_VERSION_3 = 0x20080522
+			var capData struct {
+				effective   uint32
+				permitted   uint32
+				inheritable uint32
+			}
+			if _, _, errno := syscall.Syscall(syscall.SYS_CAPGET, uintptr(unsafe.Pointer(&struct {
+				version uint32
+				pid     int32
+			}{_LINUX_CAPABILITY_VERSION_3, 0})), uintptr(unsafe.Pointer(&capData)), 0); errno != 0 {
+				return os.NewSyscallError("capget", errno)
+			}
+
+			if max(capData.effective, capData.permitted, capData.inheritable) != 0 {
+				return fmt.Errorf(
+					"effective = %d, permitted = %d, inheritable = %d",
+					capData.effective, capData.permitted, capData.inheritable,
+				)
+			}
+
 			wantHost := hostnameFromTestCase(tc.name)
 			if host, err := os.Hostname(); err != nil {
 				return fmt.Errorf("cannot get hostname: %v", err)
@@ -775,7 +889,7 @@ func TestMain(m *testing.M) {
 		}
 		c.MustParse(os.Args[1:], func(err error) {
 			if err != nil {
-				log.Fatal(err.Error())
+				log.Fatal(err)
 			}
 		})
 		return

container/init.go

@@ -11,11 +11,13 @@ import (
 	"path/filepath"
 	"slices"
 	"strconv"
+	"strings"
 	"sync"
 	"sync/atomic"
 	. "syscall"
 	"time"
 
+	"hakurei.app/check"
 	"hakurei.app/container/seccomp"
 	"hakurei.app/ext"
 	"hakurei.app/fhs"
@@ -182,23 +184,33 @@ func initEntrypoint(k syscallDispatcher, msg message.Msg) {
 		cancel()
 	}
 
+	uid, gid := param.Uid, param.Gid
+	if param.InitAsRoot {
+		uid, gid = 0, 0
+	}
+
 	// write uid/gid map here so parent does not need to set dumpable
 	if err := k.setDumpable(ext.SUID_DUMP_USER); err != nil {
 		k.fatalf(msg, "cannot set SUID_DUMP_USER: %v", err)
 	}
-	if err := k.writeFile(fhs.Proc+"self/uid_map",
+	if err := k.writeFile(
-		append([]byte{}, strconv.Itoa(param.Uid)+" "+strconv.Itoa(param.HostUid)+" 1\n"...),
+		fhs.Proc+"self/uid_map",
-		0); err != nil {
+		[]byte(strconv.Itoa(uid)+" "+strconv.Itoa(param.HostUid)+" 1\n"),
+		0,
+	); err != nil {
 		k.fatalf(msg, "%v", err)
 	}
-	if err := k.writeFile(fhs.Proc+"self/setgroups",
+	if err := k.writeFile(
+		fhs.Proc+"self/setgroups",
 		[]byte("deny\n"),
-		0); err != nil && !os.IsNotExist(err) {
+		0,
+	); err != nil && !os.IsNotExist(err) {
 		k.fatalf(msg, "%v", err)
 	}
 	if err := k.writeFile(fhs.Proc+"self/gid_map",
-		append([]byte{}, strconv.Itoa(param.Gid)+" "+strconv.Itoa(param.HostGid)+" 1\n"...),
+		[]byte(strconv.Itoa(gid)+" "+strconv.Itoa(param.HostGid)+" 1\n"),
-		0); err != nil {
+		0,
+	); err != nil {
 		k.fatalf(msg, "%v", err)
 	}
 	if err := k.setDumpable(ext.SUID_DUMP_DISABLE); err != nil {
@@ -230,6 +242,16 @@ func initEntrypoint(k syscallDispatcher, msg message.Msg) {
 		k.fatalf(msg, "cannot enter intermediate host path: %v", err)
 	}
 
+	if len(param.Binfmt) > 0 {
+		for i, e := range param.Binfmt {
+			if pathname, err := k.evalSymlinks(e.Interpreter.String()); err != nil {
+				k.fatal(msg, err)
+			} else if param.Binfmt[i].Interpreter, err = check.NewAbs(pathname); err != nil {
+				k.fatal(msg, err)
+			}
+		}
+	}
+
 	/* early is called right before pivot_root into intermediate root;
 	this step is mostly for gathering information that would otherwise be
 	difficult to obtain via library functions after pivot_root, and
@@ -285,6 +307,48 @@ func initEntrypoint(k syscallDispatcher, msg message.Msg) {
 		}
 	}
 
+	if len(param.Binfmt) > 0 {
+		const interpreter = "/interpreter"
+
+		if param.BinfmtPath == nil {
+			param.BinfmtPath = fhs.AbsProcSys.Append("fs/binfmt_misc")
+		}
+		binfmt := sysrootPath + param.BinfmtPath.String()
+		if err := k.mkdirAll(binfmt, 0); err != nil {
+			k.fatal(msg, err)
+		}
+		if err := k.mount(
+			SourceBinfmtMisc,
+			binfmt,
+			FstypeBinfmtMisc,
+			MS_NOSUID|MS_NOEXEC|MS_NODEV,
+			zeroString,
+		); err != nil {
+			k.fatal(msg, err)
+		}
+
+		var buf strings.Builder
+		buf.Grow(1920)
+
+		register := binfmt + "/register"
+		for i, e := range param.Binfmt {
+			if err := k.symlink(hostPath+e.Interpreter.String(), interpreter); err != nil {
+				k.fatal(msg, err)
+			} else if err = k.writeFile(register, []byte(":"+
+				strconv.Itoa(i)+":"+
+				"M:"+
+				strconv.Itoa(int(e.Offset))+":"+
+				escapeBinfmt(&buf, e.Magic)+":"+
+				escapeBinfmt(&buf, e.Mask)+":"+
+				interpreter+":"+
+				"F"), 0); err != nil {
+				k.fatal(msg, err)
+			} else if err = k.remove(interpreter); err != nil {
+				k.fatal(msg, err)
+			}
+		}
+	}
+
 	// setup requiring host root complete at this point
 	if err := k.mount(hostDir, hostDir, zeroString, MS_SILENT|MS_REC|MS_PRIVATE, zeroString); err != nil {
 		k.fatalf(msg, "cannot make host root rprivate: %v", optionalErrorUnwrap(err))
@@ -323,11 +387,19 @@ func initEntrypoint(k syscallDispatcher, msg message.Msg) {
 		}
 	}
 
+	var keepCaps []uintptr
+	if param.Privileged {
+		keepCaps = append(keepCaps, CAP_SYS_ADMIN, CAP_SETPCAP)
+	}
+	if param.InitAsRoot {
+		keepCaps = append(keepCaps, CAP_SETFCAP)
+	}
+
 	if err := k.capAmbientClearAll(); err != nil {
 		k.fatalf(msg, "cannot clear the ambient capability set: %v", err)
 	}
-	for i := uintptr(0); i <= lastcap; i++ {
+	for i := range lastcap + 1 {
-		if param.Privileged && i == CAP_SYS_ADMIN {
+		if slices.Contains(keepCaps, i) {
 			continue
 		}
 		if err := k.capBoundingSetDrop(i); err != nil {
@@ -336,20 +408,23 @@ func initEntrypoint(k syscallDispatcher, msg message.Msg) {
 	}
 
 	var keep [2]uint32
-	if param.Privileged {
+	for _, c := range keepCaps {
-		keep[capToIndex(CAP_SYS_ADMIN)] |= capToMask(CAP_SYS_ADMIN)
+		keep[capToIndex(c)] |= capToMask(c)
-
-		if err := k.capAmbientRaise(CAP_SYS_ADMIN); err != nil {
-			k.fatalf(msg, "cannot raise CAP_SYS_ADMIN: %v", err)
-		}
 	}
+
 	if err := k.capset(
 		&capHeader{_LINUX_CAPABILITY_VERSION_3, 0},
-		&[2]capData{{0, keep[0], keep[0]}, {0, keep[1], keep[1]}},
+		&[2]capData{{keep[0], keep[0], keep[0]}, {keep[1], keep[1], keep[1]}},
 	); err != nil {
 		k.fatalf(msg, "cannot capset: %v", err)
 	}
 
+	for _, c := range keepCaps {
+		if err := k.capAmbientRaise(c); err != nil {
+			k.fatalf(msg, "cannot raise %#x: %v", c, err)
+		}
+	}
+
 	if !param.SeccompDisable {
 		rules := param.SeccompRules
 		if len(rules) == 0 { // non-empty rules slice always overrides presets
@@ -474,6 +549,14 @@ func initEntrypoint(k syscallDispatcher, msg message.Msg) {
 	cmd.ExtraFiles = extraFiles
 	cmd.Dir = param.Dir.String()
 
+	if param.InitAsRoot {
+		cmd.SysProcAttr = &SysProcAttr{
+			Cloneflags:  CLONE_NEWUSER,
+			UidMappings: []SysProcIDMap{{ContainerID: param.Uid, HostID: 0, Size: 1}},
+			GidMappings: []SysProcIDMap{{ContainerID: param.Gid, HostID: 0, Size: 1}},
+		}
+	}
+
 	msg.Verbosef("starting initial process %s", param.Path)
 	if err := k.start(cmd); err != nil {
 		k.fatalf(msg, "%v", err)

container/init_test.go

@@ -1624,7 +1624,6 @@ func TestInitEntrypoint(t *testing.T) {
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x5)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x6)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x7)}, nil, nil),
-				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x8)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x9)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0xa)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0xb)}, nil, nil),
@@ -1656,8 +1655,9 @@ func TestInitEntrypoint(t *testing.T) {
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x26)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x27)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x28)}, nil, nil),
+				call("capset", stub.ExpectArgs{&capHeader{_LINUX_CAPABILITY_VERSION_3, 0}, &[2]capData{{0x200100, 0x200100, 0x200100}, {0, 0, 0}}}, nil, nil),
 				call("capAmbientRaise", stub.ExpectArgs{uintptr(0x15)}, nil, stub.UniqueError(19)),
-				call("fatalf", stub.ExpectArgs{"cannot raise CAP_SYS_ADMIN: %v", []any{stub.UniqueError(19)}}, nil, nil),
+				call("fatalf", stub.ExpectArgs{"cannot raise %#x: %v", []any{uintptr(0x15), stub.UniqueError(19)}}, nil, nil),
 			},
 		}, nil},
 
@@ -1731,7 +1731,6 @@ func TestInitEntrypoint(t *testing.T) {
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x5)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x6)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x7)}, nil, nil),
-				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x8)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x9)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0xa)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0xb)}, nil, nil),
@@ -1763,8 +1762,7 @@ func TestInitEntrypoint(t *testing.T) {
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x26)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x27)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x28)}, nil, nil),
-				call("capAmbientRaise", stub.ExpectArgs{uintptr(0x15)}, nil, nil),
+				call("capset", stub.ExpectArgs{&capHeader{_LINUX_CAPABILITY_VERSION_3, 0}, &[2]capData{{0x200100, 0x200100, 0x200100}, {0, 0, 0}}}, nil, stub.UniqueError(17)),
-				call("capset", stub.ExpectArgs{&capHeader{_LINUX_CAPABILITY_VERSION_3, 0}, &[2]capData{{0, 0x200000, 0x200000}, {0, 0, 0}}}, nil, stub.UniqueError(17)),
 				call("fatalf", stub.ExpectArgs{"cannot capset: %v", []any{stub.UniqueError(17)}}, nil, nil),
 			},
 		}, nil},
@@ -1839,7 +1837,6 @@ func TestInitEntrypoint(t *testing.T) {
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x5)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x6)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x7)}, nil, nil),
-				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x8)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x9)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0xa)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0xb)}, nil, nil),
@@ -1871,8 +1868,9 @@ func TestInitEntrypoint(t *testing.T) {
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x26)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x27)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x28)}, nil, nil),
+				call("capset", stub.ExpectArgs{&capHeader{_LINUX_CAPABILITY_VERSION_3, 0}, &[2]capData{{0x200100, 0x200100, 0x200100}, {0, 0, 0}}}, nil, nil),
 				call("capAmbientRaise", stub.ExpectArgs{uintptr(0x15)}, nil, nil),
-				call("capset", stub.ExpectArgs{&capHeader{_LINUX_CAPABILITY_VERSION_3, 0}, &[2]capData{{0, 0x200000, 0x200000}, {0, 0, 0}}}, nil, nil),
+				call("capAmbientRaise", stub.ExpectArgs{uintptr(0x8)}, nil, nil),
 				call("verbosef", stub.ExpectArgs{"resolving presets %#x", []any{std.FilterPreset(0xf)}}, nil, nil),
 				call("seccompLoad", stub.ExpectArgs{seccomp.Preset(0xf, 0), seccomp.ExportFlag(0)}, nil, stub.UniqueError(15)),
 				call("fatalf", stub.ExpectArgs{"cannot load syscall filter: %v", []any{stub.UniqueError(15)}}, nil, nil),
@@ -2699,7 +2697,6 @@ func TestInitEntrypoint(t *testing.T) {
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x5)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x6)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x7)}, nil, nil),
-				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x8)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x9)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0xa)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0xb)}, nil, nil),
@@ -2731,8 +2728,9 @@ func TestInitEntrypoint(t *testing.T) {
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x26)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x27)}, nil, nil),
 				call("capBoundingSetDrop", stub.ExpectArgs{uintptr(0x28)}, nil, nil),
+				call("capset", stub.ExpectArgs{&capHeader{_LINUX_CAPABILITY_VERSION_3, 0}, &[2]capData{{0x200100, 0x200100, 0x200100}, {0, 0, 0}}}, nil, nil),
 				call("capAmbientRaise", stub.ExpectArgs{uintptr(0x15)}, nil, nil),
-				call("capset", stub.ExpectArgs{&capHeader{_LINUX_CAPABILITY_VERSION_3, 0}, &[2]capData{{0, 0x200000, 0x200000}, {0, 0, 0}}}, nil, nil),
+				call("capAmbientRaise", stub.ExpectArgs{uintptr(0x8)}, nil, nil),
 				call("verbosef", stub.ExpectArgs{"resolving presets %#x", []any{std.FilterPreset(0xf)}}, nil, nil),
 				call("seccompLoad", stub.ExpectArgs{seccomp.Preset(0xf, 0), seccomp.ExportFlag(0)}, nil, nil),
 				call("verbosef", stub.ExpectArgs{"%d filter rules loaded", []any{73}}, nil, nil),

container/mount.go

@@ -40,6 +40,9 @@ const (
 	// SourceMqueue is used when mounting mqueue.
 	// Note that any source value is allowed when fstype is [FstypeMqueue].
 	SourceMqueue = "mqueue"
+	// SourceBinfmtMisc is used when mounting binfmt_misc.
+	// Note that any source value is allowed when fstype is [SourceBinfmtMisc].
+	SourceBinfmtMisc = "binfmt_misc"
 	// SourceOverlay is used when mounting overlay.
 	// Note that any source value is allowed when fstype is [FstypeOverlay].
 	SourceOverlay = "overlay"
@@ -70,6 +73,9 @@ const (
 	// FstypeMqueue represents the mqueue pseudo-filesystem.
 	// This filesystem type is usually mounted on /dev/mqueue.
 	FstypeMqueue = "mqueue"
+	// FstypeBinfmtMisc represents the binfmt_misc pseudo-filesystem.
+	// This filesystem type is usually mounted on /proc/sys/fs/binfmt_misc.
+	FstypeBinfmtMisc = "binfmt_misc"
 	// FstypeOverlay represents the overlay pseudo-filesystem.
 	// This filesystem type can be mounted anywhere in the container filesystem.
 	FstypeOverlay = "overlay"

fhs/abs.go

@@ -42,6 +42,8 @@ var (
 	AbsDevShm = unsafeAbs(DevShm)
 	// AbsProc is [Proc] as [check.Absolute].
 	AbsProc = unsafeAbs(Proc)
+	// AbsProcSys is [ProcSys] as [check.Absolute].
+	AbsProcSys = unsafeAbs(ProcSys)
 	// AbsProcSelfExe is [ProcSelfExe] as [check.Absolute].
 	AbsProcSelfExe = unsafeAbs(ProcSelfExe)
 	// AbsSys is [Sys] as [check.Absolute].

internal/pkg/exec.go

@@ -631,12 +631,6 @@ func (a *execArtifact) cure(f *FContext, hostNet bool) (err error) {
 			_ = stdout.Close()
 			return
 		}
-		defer func() {
-			if err != nil && !errors.As(err, new(*exec.ExitError)) {
-				_ = stdout.Close()
-				_ = stderr.Close()
-			}
-		}()
 
 		brStdout, brStderr := f.cache.getReader(stdout), f.cache.getReader(stderr)
 		stdoutDone, stderrDone := make(chan struct{}), make(chan struct{})
@@ -651,6 +645,11 @@ func (a *execArtifact) cure(f *FContext, hostNet bool) (err error) {
 			io.TeeReader(brStderr, status),
 		)
 		defer func() {
+			if err != nil && !errors.As(err, new(*exec.ExitError)) {
+				_ = stdout.Close()
+				_ = stderr.Close()
+			}
+
 			<-stdoutDone
 			<-stderrDone
 			f.cache.putReader(brStdout)

internal/pkg/pkg_test.go

@@ -423,7 +423,7 @@ func checkWithCache(t *testing.T, testCases []cacheTestCase) {
 			msg := message.New(log.New(os.Stderr, "cache: ", 0))
 			msg.SwapVerbose(testing.Verbose())
 
-			flags := tc.flags
+			flags := tc.flags | pkg.CSuppressInit
 
 			if info.CanDegrade {
 				if _, err := landlock.GetABI(); err != nil {
@@ -544,7 +544,11 @@ func cureMany(t *testing.T, c *pkg.Cache, steps []cureStep) {
 			t.Fatalf("Cure: pathname = %q, want %q", pathname, step.pathname)
 		} else if step.output == nil || checksum != makeChecksumH(step.output.hash()) {
 			if pathname != nil {
-				t.Fatal(expectsFrom(pathname.String()))
+				if name, _err := filepath.EvalSymlinks(pathname.String()); _err != nil {
+					t.Fatal(_err)
+				} else {
+					t.Fatal(expectsFrom(name))
+				}
 			} else if checksum != (unique.Handle[pkg.Checksum]{}) {
 				t.Fatalf("Cure: unexpected checksum %s", pkg.Encode(checksum.Value()))
 			}