The do_remount function in fs/namespace.c in the Linux kernel through 3.16.1 does not maintain the MNT_LOCK_READONLY bit across a remount of a bind mount, which allows local users to bypass an intended read-only restriction and defeat certain sandbox protection mechanisms via a "mount -o remount" command within a user namespace.
https://github.com/torvalds/linux/commit/a6138db815df5ee542d848318e5dae681590fccd
https://bugzilla.redhat.com/show_bug.cgi?id=1129662
http://www.ubuntu.com/usn/USN-2318-1
http://www.ubuntu.com/usn/USN-2317-1
http://www.securityfocus.com/bid/69214