In a container, try to register binfmt_misc

If we're being started up in a container, default to attempting to
register any emulators for which we find configurations of the type
included in Fedora's qemu-user-static packages.

Use a hidden global --register-binfmt flag to control this, and set its
default based on whether or not $container is set to "oci".

Signed-off-by: Nalin Dahyabhai <[email protected]>

cmd/buildah/main.go

@@ -12,6 +12,7 @@ import (
 
 	"github.com/containers/buildah"
 	"github.com/containers/buildah/define"
+	"github.com/containers/buildah/pkg/binfmt"
 	"github.com/containers/buildah/pkg/cli"
 	"github.com/containers/buildah/pkg/parse"
 	"github.com/containers/common/pkg/config"
@@ -40,6 +41,7 @@ type globalFlags struct {
 	MemoryProfile              string
 	UserShortNameAliasConfPath string
 	CgroupManager              string
+	registerBinfmt             bool
 }
 
 var rootCmd = &cobra.Command{
@@ -118,6 +120,10 @@ func init() {
 	if err := rootCmd.PersistentFlags().MarkHidden("memory-profile"); err != nil {
 		logrus.Fatalf("unable to mark memory-profile flag as hidden: %v", err)
 	}
+	rootCmd.PersistentFlags().BoolVar(&globalFlagResults.registerBinfmt, "register-binfmt", os.Getenv("container") != "", "attempt to register binfmt_misc interpreters")
+	if err := rootCmd.PersistentFlags().MarkHidden("register-binfmt"); err != nil {
+		logrus.Fatalf("unable to mark register-binfmt flag as hidden: %v", err)
+	}
 }
 
 func initConfig() {
@@ -147,6 +153,11 @@ func before(cmd *cobra.Command) error {
 	}
 	debugCapabilities()
 	unshare.MaybeReexecUsingUserNamespace(false)
+	if globalFlagResults.registerBinfmt {
+		if err := binfmt.Register(nil); err != nil {
+			logrus.Debugf("registering binfmt_misc interpreters: %v", err)
+		}
+	}
 	if globalFlagResults.CPUProfile != "" {
 		globalFlagResults.cpuProfileFile, err = os.Create(globalFlagResults.CPUProfile)
 		if err != nil {

pkg/binfmt/binfmt.go

@@ -0,0 +1,83 @@
+//go:build linux
+
+package binfmt
+
+import (
+	"bufio"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"syscall"
+
+	"github.com/containers/storage/pkg/unshare"
+	"github.com/sirupsen/logrus"
+	"golang.org/x/sys/unix"
+)
+
+// MaybeRegister() calls Register() if the current context is a rootless one,
+// or if the "container" environment variable suggests that we're in a
+// container.
+func MaybeRegister(configurationSearchDirectories []string) error {
+	if unshare.IsRootless() || os.Getenv("container") != "" { // we _also_ own our own mount namespace
+		return Register(configurationSearchDirectories)
+	}
+	return nil
+}
+
+// Register() registers binfmt.d emulators described by configuration files in
+// the passed-in slice of directories, or in the union of /etc/binfmt.d,
+// /run/binfmt.d, and /usr/lib/binfmt.d if the slice has no items.  If any
+// emulators are configured, it will attempt to mount a binfmt_misc filesystem
+// in the current mount namespace first, ignoring only EPERM and EACCES errors.
+func Register(configurationSearchDirectories []string) error {
+	if len(configurationSearchDirectories) == 0 {
+		configurationSearchDirectories = []string{"/etc/binfmt.d", "/run/binfmt.d", "/usr/lib/binfmt.d"}
+	}
+	mounted := false
+	for _, searchDir := range configurationSearchDirectories {
+		globs, err := filepath.Glob(filepath.Join(searchDir, "*.conf"))
+		if err != nil {
+			return fmt.Errorf("looking for binfmt.d configuration in %q: %w", searchDir, err)
+		}
+		for _, conf := range globs {
+			f, err := os.Open(conf)
+			if err != nil {
+				return fmt.Errorf("reading binfmt.d configuration: %w", err)
+			}
+			scanner := bufio.NewScanner(f)
+			for scanner.Scan() {
+				line := strings.TrimSpace(scanner.Text())
+				if len(line) == 0 || line[0] == ';' || line[0] == '#' {
+					continue
+				}
+				if !mounted {
+					if err = unix.Mount("none", "/proc/sys/fs/binfmt_misc", "binfmt_misc", 0, ""); err != nil {
+						if errors.Is(err, syscall.EPERM) || errors.Is(err, syscall.EACCES) {
+							// well, we tried. no need to make a stink about it
+							return nil
+						}
+						return fmt.Errorf("mounting binfmt_misc: %w", err)
+					}
+					mounted = true
+				}
+				reg, err := os.Create("/proc/sys/fs/binfmt_misc/register")
+				if err != nil {
+					return fmt.Errorf("registering(open): %w", err)
+				}
+				if _, err = fmt.Fprintf(reg, "%s\n", line); err != nil {
+					return fmt.Errorf("registering(write): %w", err)
+				}
+				logrus.Tracef("registered binfmt %q", line)
+				if err = reg.Close(); err != nil {
+					return fmt.Errorf("registering(close): %w", err)
+				}
+			}
+			if err := f.Close(); err != nil {
+				return fmt.Errorf("reading binfmt.d configuration: %w", err)
+			}
+		}
+	}
+	return nil
+}

pkg/binfmt/binfmt_unsupported.go

@@ -0,0 +1,15 @@
+//go:build !linux
+
+package binfmt
+
+import "syscall"
+
+// MaybeRegister() returns no error.
+func MaybeRegister(configurationSearchDirectories []string) error {
+	return nil
+}
+
+// Register() returns an error.
+func Register(configurationSearchDirectories []string) error {
+	return syscall.ENOSYS
+}