commit 380cf5ba6b0a ("fs: Treat foreign mounts as nosuid"), in addition to
the intended effect, also prevented suid execution of memfd files.
(And I think that's a good change.)
Signed-off-by: Jann Horn <jannh@xxxxxxxxxx>
---
fs/namespace.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/fs/namespace.c b/fs/namespace.c
index a7f91265ea67..e68488426e42 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -3400,6 +3400,9 @@ bool mnt_may_suid(struct vfsmount *mnt)
* prevents namespaces from trusting potentially unsafe
* suid/sgid bits, file caps, or security labels that originate
* in other namespaces.
+ *
+ * check_mnt() rejects MNT_NS_INTERNAL mounts; this means that suid
+ * execution is blocked for files on internal mounts, e.g. memfds.
*/
return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
current_in_userns(mnt->mnt_sb->s_user_ns);
--
2.20.0.rc0.387.gc7a69e6b6c-goog
