@terdon (re. /proc/sys/fs/protected_*) well the kernel does all the other access controls too, and doing that in the kernel makes it immediately available to all tools. More to the point, the kernel is the only place where those checks can be done without races. The vulnerability being thwarted there is something like putting a symlink in /tmp, and then tricking a privileged process into following it (modifying /etc/passwd or whatever).
If the privileged process tries to check it first, the attacker can still race to change an allowed file to a symlink, and they might get to try again and again until it works. (Sorry if I'm repeating what you already know, I just noticed the linked article didn't seem to explain it too much.)