On Wed, 15 Aug 2007, Wojciech Purczynski wrote:
> Sending a signal to privileged process is a privilege itself.
Sure. But once control was transferred to some other code that we have no
control over, we have no more control over when the signal is sent. We just
can't send that signal at arbitrary moment. If you as an attacker can create
arbitrary setuid root binary in the system, this game is not worth anymore,
since you already won.
> Under some
> circumstances this may lead to other consequences. For example I was able
> to code local root exploit using some very common suid binary, although
> its usage is somewhat limited.
>
Again, if an attacker can create arbitrary setuid root binary in the system,
the latter is already broken. And if the setuid root binary allows arbitrary
code execution on getting some signal, it is vulnerable itself. Unexpected
signals can be generated at any time by terminals, init, various system
daemons, non-blocking i/o for files open before exec(), alarms, failed pageins,
etc. If the program (not necessarily setuid/setgid) can't properly handle those
situations, it is broken by design. Signals are not something we can trust.
--
Sincerely Your, Dan.
