Re: [PATCH v3 2.6.39-rc1-tip 12/26] 12: uprobes: slot allocation for uprobes

[Stephen Smalley <sds@tycho.nsa.gov>]

On Wed, 2011-04-20 at 10:51 -0400, Frank Ch. Eigler wrote:
> eparis wrote:
> 
> > [...]
> > Now how to fix the problems you were seeing.  If you run a modern
> > system with a GUI I'm willing to bet the pop-up window told you
> > exactly how to fix your problem.  [...]
> >
> > 1) chcon -t unconfined_execmem_t /path/to/your/binary
> > 2) setsebool -P allow_execmem 1
> > [...]
> > I believe there was a question about how JIT's work with SELinux
> > systems.  They work mostly by method #1.
> 
> Actually, that's a solution to a different problem.  Here, it's not
> particular /path/to/your/binaries that want/need selinux provileges.
> It's a kernel-driven debugging facility that needs it temporarily for
> arbitrary processes.
> 
> It's not like JITs, with known binary names.  It's not like GDB, which
> simply overwrites existing instructions in the text segment.  To make
> uprobes work fast (single-step-out-of-line), one needs one or emore
> temporary pages with unusual mapping permissions.

I would expect that (2) would solve it, but couldn't distinguish the
kernel-created mappings from userspace doing the same thing.
Alternatively, you could temporarily switch your credentials around the
mapping operation, e.g.:
old_cred = override_creds(&init_cred);
do_mmap_pgoff(...);
revert_creds(old_cred);

devtmpfs does something similar to avoid triggering permission checks on
userspace when it is internally creating and deleting nodes.

How is this ability to use this facility controlled?

-- 
Stephen Smalley
National Security Agency