Entrypoint acts as root on nothing outside /data (closes #80)

`bin/entrypoint.sh` now runs `netwatch-server prepare-data-dir`, which
refuses a `DATA_DIR` that is not `/data` or a path below it written in
full, then creates `DATA_DIR`, gives `/data` and everything in it to
`netwatch`, and sets mode 750 on `/data` and `DATA_DIR`. Every step goes
through a Go `os.Root` opened on `/data`, and the modes are set on the
opened directories rather than by name, so neither a symbolic link
already there nor one a host process swaps in while the container
starts can make root create or change anything outside `/data`. The
README says which `DATA_DIR` values are accepted.

Model: opus-5-5

backend/README.md

@@ -107,8 +107,9 @@ this server. The image's entrypoint, `bin/entrypoint.sh`, starts the server as
 user `netwatch` (uid 1000) with `BIND_ADDRESS=127.0.0.1` and `PORT=8081`, so
 only nginx reaches it, and with `TRUSTED_PROXIES=127.0.0.1/32`, so it takes the
 client address nginx passes on and no other. `DATA_DIR` is `/data/reports`, on
-the `/data` volume; the entrypoint creates it and gives it and `/data` to
-`netwatch` before starting the server. nginx replaces the security headers
+the `/data` volume; before starting the server, the entrypoint creates it and
+gives it and `/data` to `netwatch` with `netwatch-server prepare-data-dir`,
+which acts on nothing outside `/data`. nginx replaces the security headers
 this server sets with those in the root `security-headers.conf`, so those are
 what clients of the image see.
 

backend/cmd/netwatch-server/main.go

@@ -4,6 +4,7 @@ package main
 import (
 	"fmt"
 	"os"
+	"os/user"
 
 	"sneak.berlin/go/netwatch/internal/config"
 	"sneak.berlin/go/netwatch/internal/globals"
@@ -37,6 +38,26 @@ func main() {
 		return
 	}
 
+	// "netwatch-server prepare-data-dir DATA_DIR" gets DATA_DIR ready
+	// for the netwatch user, or exits 1 with the error; see
+	// reportbuf.PrepareDataDir. bin/entrypoint.sh runs it as root
+	// before it starts this server as that user.
+	if len(os.Args) == 3 && os.Args[1] == "prepare-data-dir" {
+		netwatch, err := user.Lookup("netwatch")
+		if err != nil {
+			fmt.Fprintln(os.Stderr, err)
+			os.Exit(1)
+		}
+
+		err = reportbuf.PrepareDataDir("/data", os.Args[2], netwatch)
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "DATA_DIR '%s': %v\n", os.Args[2], err)
+			os.Exit(1)
+		}
+
+		return
+	}
+
 	globals.Appname = Appname
 	globals.Version = Version
 

backend/internal/reportbuf/datadir.go

@@ -0,0 +1,150 @@
+package reportbuf
+
+import (
+	"errors"
+	"io/fs"
+	"os"
+	"os/user"
+	"path/filepath"
+	"strconv"
+	"syscall"
+)
+
+// ErrDataDirOutsideVolume is returned by PrepareDataDir for a DATA_DIR
+// that is not the volume or a path below it, written in full.
+var ErrDataDirOutsideVolume = errors.New(
+	"must be /data or a path below it, with no '.', '..' or extra '/'")
+
+// PrepareDataDir gets dir, the server's DATA_DIR, ready for owner, the
+// user the server runs as, so that a host directory mounted at volume,
+// /data in the image, needs no preparing: it creates dir, gives volume
+// and everything in it to owner, and gives volume and dir the mode the
+// server gives a directory it creates. dir must be volume or a path
+// below it, with no '.', '..', empty part or '/' at the end.
+//
+// bin/entrypoint.sh runs this as root, which would follow a symbolic
+// link anywhere, so every step goes through an os.Root opened on
+// volume: it follows a link only when it is written as a relative
+// path that stays inside volume, and refuses any other. A process on
+// the host can swap a link onto a path in volume at any moment while
+// this runs. Even then, the os.Root checks each link as it reaches
+// it. MkdirAll creates each directory inside a parent it already has
+// open, never following a link at the name it creates, and follows a
+// link on the path only as the os.Root allows, so a relative link
+// inside volume can lead it to create directories elsewhere inside
+// volume. Lchown never changes what a link points to, and the modes
+// are set on directories already opened (see chmodDir), so the most
+// that process can do is make a step fail or wait, or act on
+// something else inside volume.
+func PrepareDataDir(volume, dir string, owner *user.User) error {
+	// rel is dir as a path from volume; IsLocal is false for one that
+	// leads out of it.
+	rel, err := filepath.Rel(volume, dir)
+	if err != nil || dir != filepath.Clean(dir) || !filepath.IsLocal(rel) {
+		return ErrDataDirOutsideVolume
+	}
+
+	uid, err := strconv.Atoi(owner.Uid)
+	if err != nil {
+		return err
+	}
+
+	gid, err := strconv.Atoi(owner.Gid)
+	if err != nil {
+		return err
+	}
+
+	root, err := os.OpenRoot(volume)
+	if err != nil {
+		return err
+	}
+
+	defer func() { _ = root.Close() }()
+
+	err = root.MkdirAll(rel, dirPerms)
+	if err != nil {
+		return err
+	}
+
+	err = lchownAll(root, ".", uid, gid)
+	if err != nil {
+		return err
+	}
+
+	err = chmodDir(root, ".")
+	if err != nil {
+		return err
+	}
+
+	return chmodDir(root, rel)
+}
+
+// lchownAll gives name, a directory inside root, and everything in it
+// to uid and gid. It reads each directory opened through root, not
+// through root.FS(), which refuses a name that is not valid UTF-8, and
+// calls Lchown on every entry, which gives a symbolic link itself to
+// them, not what it points to. It goes into an entry only when the
+// read found a directory there, so it follows no link it finds; one
+// swapped in for that directory afterwards is followed only as the
+// os.Root allows.
+func lchownAll(root *os.Root, name string, uid, gid int) error {
+	err := root.Lchown(name, uid, gid)
+	if err != nil {
+		return err
+	}
+
+	dir, err := root.Open(name)
+	if err != nil {
+		return err
+	}
+
+	entries, err := dir.ReadDir(-1)
+	_ = dir.Close()
+
+	if err != nil {
+		return err
+	}
+
+	for _, entry := range entries {
+		entryName := filepath.Join(name, entry.Name())
+		if entry.IsDir() {
+			err = lchownAll(root, entryName, uid, gid)
+		} else {
+			err = root.Lchown(entryName, uid, gid)
+		}
+
+		if err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+// chmodDir gives name, a directory inside root, the mode the server
+// gives a directory it creates. Root.Chmod would not hold: on Linux it
+// checks that name is not a symbolic link, then sets the mode by name,
+// following a link swapped in between. So chmodDir opens name through
+// root and sets the mode on the open directory. It refuses anything
+// but a directory: a directory has no second name (hard link), so the
+// one opened is inside root, where any other file could be a hard link
+// to one outside.
+func chmodDir(root *os.Root, name string) error {
+	dir, err := root.Open(name)
+	if err != nil {
+		return err
+	}
+
+	defer func() { _ = dir.Close() }()
+
+	info, err := dir.Stat()
+	if err != nil {
+		return err
+	}
+
+	if !info.IsDir() {
+		return &fs.PathError{Op: "chmod", Path: name, Err: syscall.ENOTDIR}
+	}
+
+	return dir.Chmod(dirPerms)
+}

backend/internal/reportbuf/datadir_test.go

@@ -0,0 +1,307 @@
+package reportbuf_test
+
+import (
+	"errors"
+	"io/fs"
+	"os"
+	"os/user"
+	"path/filepath"
+	"strconv"
+	"syscall"
+	"testing"
+
+	"sneak.berlin/go/netwatch/internal/reportbuf"
+)
+
+// reports is the last part of DATA_DIR in these tests, as in the
+// image's /data/reports.
+const reports = "reports"
+
+// currentUser is the user the test runs as, the only owner a test not
+// run as root can give files to.
+func currentUser() *user.User {
+	return &user.User{
+		Uid: strconv.Itoa(os.Getuid()),
+		Gid: strconv.Itoa(os.Getgid()),
+	}
+}
+
+// tempDirMode700 is a new directory in a t.TempDir with mode 0700, so
+// a test can tell that PrepareDataDir left its mode alone.
+func tempDirMode700(t *testing.T) string {
+	t.Helper()
+
+	dir := filepath.Join(t.TempDir(), "d")
+
+	err := os.Mkdir(dir, 0o700)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	return dir
+}
+
+func requireMode(t *testing.T, path string, want fs.FileMode) {
+	t.Helper()
+
+	info, err := os.Stat(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if info.Mode() != want {
+		t.Errorf("%s: mode %v, want %v", path, info.Mode(), want)
+	}
+}
+
+func requireMissing(t *testing.T, path string) {
+	t.Helper()
+
+	_, err := os.Lstat(path)
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Errorf("%s: created, or Lstat failed: %v", path, err)
+	}
+}
+
+func requireOwner(t *testing.T, path string, uid, gid uint32) {
+	t.Helper()
+
+	info, err := os.Lstat(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	stat, _ := info.Sys().(*syscall.Stat_t)
+	if stat.Uid != uid || stat.Gid != gid {
+		t.Errorf("%s: owner %d:%d, want %d:%d", path, stat.Uid, stat.Gid,
+			uid, gid)
+	}
+}
+
+func TestPrepareDataDirCreatesDataDir(t *testing.T) {
+	t.Parallel()
+
+	volume := t.TempDir()
+	dir := filepath.Join(volume, "a", reports)
+
+	err := reportbuf.PrepareDataDir(volume, dir, currentUser())
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	requireMode(t, volume, fs.ModeDir|0o750)
+	requireMode(t, dir, fs.ModeDir|0o750)
+}
+
+// TestPrepareDataDirSetsModeOfExistingDataDir: a DATA_DIR already on
+// the host with another mode gets the mode too, not only a new one.
+func TestPrepareDataDirSetsModeOfExistingDataDir(t *testing.T) {
+	t.Parallel()
+
+	volume := t.TempDir()
+	dir := filepath.Join(volume, reports)
+
+	err := os.Mkdir(dir, 0o700)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = reportbuf.PrepareDataDir(volume, dir, currentUser())
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	requireMode(t, dir, fs.ModeDir|0o750)
+}
+
+func TestPrepareDataDirTakesTheVolumeItself(t *testing.T) {
+	t.Parallel()
+
+	volume := t.TempDir()
+
+	err := reportbuf.PrepareDataDir(volume, volume, currentUser())
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	requireMode(t, volume, fs.ModeDir|0o750)
+}
+
+// TestPrepareDataDirRefusesDataDirOutsideVolume covers a DATA_DIR that
+// is relative, outside the volume, or not written in full.
+func TestPrepareDataDirRefusesDataDirOutsideVolume(t *testing.T) {
+	t.Parallel()
+
+	volume := tempDirMode700(t)
+	for _, dir := range []string{
+		reports, volume + "/../new", volume + "/", volume + "//" + reports,
+		volume + "/./" + reports, volume + "/" + reports + "/..", volume + "x",
+		"/etc",
+	} {
+		err := reportbuf.PrepareDataDir(volume, dir, currentUser())
+		if !errors.Is(err, reportbuf.ErrDataDirOutsideVolume) {
+			t.Errorf("%q: error = %v, want ErrDataDirOutsideVolume", dir, err)
+		}
+	}
+
+	requireMissing(t, filepath.Join(filepath.Dir(volume), "new"))
+	requireMode(t, volume, fs.ModeDir|0o700)
+}
+
+// TestPrepareDataDirRefusesLinkOutOfVolume puts a symbolic link to a
+// directory outside the volume on the path to DATA_DIR, written as a
+// full path and as one that climbs out with '..', and as DATA_DIR
+// itself, where the mode would be set through it.
+func TestPrepareDataDirRefusesLinkOutOfVolume(t *testing.T) {
+	t.Parallel()
+
+	for _, tc := range []struct {
+		name      string
+		climbsOut bool
+		link, dir string
+	}{
+		{"full path", false, "x", "x/reports"},
+		{"climbs out", true, "x", "x/reports"},
+		{"DATA_DIR itself", false, reports, reports},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			outside := tempDirMode700(t)
+			volume := t.TempDir()
+
+			climbOut, err := filepath.Rel(volume, outside)
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			target := outside
+			if tc.climbsOut {
+				target = climbOut
+			}
+
+			err = os.Symlink(target, filepath.Join(volume, tc.link))
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			err = reportbuf.PrepareDataDir(volume,
+				filepath.Join(volume, tc.dir), currentUser())
+			if err == nil {
+				t.Error("no error")
+			}
+
+			requireMissing(t, filepath.Join(outside, reports))
+			requireMode(t, outside, fs.ModeDir|0o700)
+		})
+	}
+}
+
+// TestPrepareDataDirRefusesDanglingLink: DATA_DIR is a symbolic link
+// to a name in the volume that does not exist, which is not created.
+func TestPrepareDataDirRefusesDanglingLink(t *testing.T) {
+	t.Parallel()
+
+	volume := t.TempDir()
+
+	err := os.Symlink("missing", filepath.Join(volume, reports))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = reportbuf.PrepareDataDir(volume, filepath.Join(volume, reports),
+		currentUser())
+	if err == nil {
+		t.Error("no error")
+	}
+
+	requireMissing(t, filepath.Join(volume, "missing"))
+}
+
+// TestPrepareDataDirTakesDirectoryNamedInLatin1: a host directory can
+// hold names that are not valid UTF-8, here "café" written in Latin-1.
+// A test not run as root can only check that PrepareDataDir goes into
+// such a directory and gives it, and what it holds, to the current
+// user.
+func TestPrepareDataDirTakesDirectoryNamedInLatin1(t *testing.T) {
+	t.Parallel()
+
+	volume := t.TempDir()
+	latin1 := filepath.Join(volume, "caf\xe9")
+
+	err := os.Mkdir(latin1, 0o700)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = os.WriteFile(filepath.Join(latin1, "f"), nil, 0o600)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	owner := currentUser()
+
+	err = reportbuf.PrepareDataDir(volume, filepath.Join(volume, reports),
+		owner)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	uid, _ := strconv.ParseUint(owner.Uid, 10, 32)
+	gid, _ := strconv.ParseUint(owner.Gid, 10, 32)
+
+	requireOwner(t, latin1, uint32(uid), uint32(gid))
+	requireOwner(t, filepath.Join(latin1, "f"), uint32(uid), uint32(gid))
+}
+
+// TestPrepareDataDirGivesVolumeToOwner gives everything in the volume
+// to a uid and gid that own nothing, which only root can do. A
+// symbolic link in the volume to a directory outside it is given to
+// them itself; what it points to is left as it was.
+func TestPrepareDataDirGivesVolumeToOwner(t *testing.T) {
+	t.Parallel()
+
+	if os.Geteuid() != 0 {
+		t.Skip("only root can give files to another uid")
+	}
+
+	outside := t.TempDir()
+	volume := t.TempDir()
+	old := filepath.Join(volume, "old")
+
+	err := os.WriteFile(filepath.Join(outside, "f"), nil, 0o600)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = os.Mkdir(old, 0o700)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = os.WriteFile(filepath.Join(old, "f"), nil, 0o600)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = os.Symlink(outside, filepath.Join(old, "link"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	err = reportbuf.PrepareDataDir(volume, filepath.Join(volume, reports),
+		&user.User{Uid: "4242", Gid: "4343"})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	for _, path := range []string{
+		volume, filepath.Join(volume, reports), old,
+		filepath.Join(old, "f"), filepath.Join(old, "link"),
+	} {
+		requireOwner(t, path, 4242, 4343)
+	}
+
+	requireOwner(t, outside, 0, 0)
+	requireOwner(t, filepath.Join(outside, "f"), 0, 0)
+}