Create the cgroupfs mount point in sysfs.

Create the /sys/fs/cgroup directory when cgroups are available. This
creates the empty directory to serve as the mountpoint, actually
mounting cgroups is left to the launcher/userspace. This is consistent
with Linux behaviour.

Without this mountpoint, getdents(2) on /sys/fs indicates an empty
directory even if the launcher mounts cgroupfs at /sys/fs/cgroup. The
launcher can't create the mountpoint directory since sysfs doesn't
support mkdir.

PiperOrigin-RevId: 398596698

2 files changed, 27 insertions, 1 deletions

diff --git a/pkg/sentry/fsimpl/sys/sys.go b/pkg/sentry/fsimpl/sys/sys.go
index f322d2747..7fcb2d26b 100644
--- a/pkg/sentry/fsimpl/sys/sys.go
+++ b/pkg/sentry/fsimpl/sys/sys.go
@@ -84,6 +84,18 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
 	fs.MaxCachedDentries = maxCachedDentries
 	fs.VFSFilesystem().Init(vfsObj, &fsType, fs)
 
+	k := kernel.KernelFromContext(ctx)
+	fsDirChildren := make(map[string]kernfs.Inode)
+	// Create an empty directory to serve as the mount point for cgroupfs when
+	// cgroups are available. This emulates Linux behaviour, see
+	// kernel/cgroup.c:cgroup_init(). Note that in Linux, userspace (typically
+	// the init process) is ultimately responsible for actually mounting
+	// cgroupfs, but the kernel creates the mountpoint. For the sentry, the
+	// launcher mounts cgroupfs.
+	if k.CgroupRegistry() != nil {
+		fsDirChildren["cgroup"] = fs.newDir(ctx, creds, defaultSysDirMode, nil)
+	}
+
 	root := fs.newDir(ctx, creds, defaultSysDirMode, map[string]kernfs.Inode{
 		"block": fs.newDir(ctx, creds, defaultSysDirMode, nil),
 		"bus":   fs.newDir(ctx, creds, defaultSysDirMode, nil),
@@ -97,7 +109,7 @@ func (fsType FilesystemType) GetFilesystem(ctx context.Context, vfsObj *vfs.Virt
 			}),
 		}),
 		"firmware": fs.newDir(ctx, creds, defaultSysDirMode, nil),
-		"fs":       fs.newDir(ctx, creds, defaultSysDirMode, nil),
+		"fs":       fs.newDir(ctx, creds, defaultSysDirMode, fsDirChildren),
 		"kernel":   kernelDir(ctx, fs, creds),
 		"module":   fs.newDir(ctx, creds, defaultSysDirMode, nil),
 		"power":    fs.newDir(ctx, creds, defaultSysDirMode, nil),
diff --git a/pkg/sentry/fsimpl/sys/sys_test.go b/pkg/sentry/fsimpl/sys/sys_test.go
index 0a0d914cc..0c46a3a13 100644
--- a/pkg/sentry/fsimpl/sys/sys_test.go
+++ b/pkg/sentry/fsimpl/sys/sys_test.go
@@ -87,3 +87,17 @@ func TestSysRootContainsExpectedEntries(t *testing.T) {
 		"power":    linux.DT_DIR,
 	})
 }
+
+func TestCgroupMountpointExists(t *testing.T) {
+	// Note: The mountpoint is only created if cgroups are available. This is
+	// the VFS2 implementation of sysfs and the test runs with VFS2 enabled, so
+	// we expect to see the mount point unconditionally.
+	s := newTestSystem(t)
+	defer s.Destroy()
+	pop := s.PathOpAtRoot("/fs")
+	s.AssertAllDirentTypes(s.ListDirents(pop), map[string]testutil.DirentType{
+		"cgroup": linux.DT_DIR,
+	})
+	pop = s.PathOpAtRoot("/fs/cgroup")
+	s.AssertAllDirentTypes(s.ListDirents(pop), map[string]testutil.DirentType{ /*empty*/ })
+}