vfs: Don't copy mount bind mounts of /proc/<pid>/ns/mnt between namespaces

Don't copy bind mounts of /proc/<pid>/ns/mnt between namespaces.
These files hold references to a mount namespace and copying them
between namespaces could result in a reference counting loop.

The current mnt_ns_loop test prevents loops on the assumption that
mounts don't cross between namespaces.  Unfortunately unsharing a
mount namespace and shared substrees can both cause mounts to
propogate between mount namespaces.

Add two flags CL_COPY_UNBINDABLE and CL_COPY_MNT_NS_FILE are added to
control this behavior, and CL_COPY_ALL is redefined as both of them.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>

1 files changed, 4 insertions, 1 deletions

diff --git a/fs/pnode.h b/fs/pnode.h
index b091445c1c4a..59e7eda1851e 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -19,11 +19,14 @@
 
 #define CL_EXPIRE               0x01
 #define CL_SLAVE                0x02
-#define CL_COPY_ALL             0x04
+#define CL_COPY_UNBINDABLE      0x04
 #define CL_MAKE_SHARED          0x08
 #define CL_PRIVATE              0x10
 #define CL_SHARED_TO_SLAVE      0x20
 #define CL_UNPRIVILEGED         0x40
+#define CL_COPY_MNT_NS_FILE     0x80
+
+#define CL_COPY_ALL             (CL_COPY_UNBINDABLE | CL_COPY_MNT_NS_FILE)
 
 static inline void set_mnt_shared(struct mount *mnt)
 {