Re: useless kernel.maps_protect and more

[Kees Cook <kees@outflux.net>]

Hi,

On Wed, Jul 16, 2008 at 06:11:35AM +0400, Alexey Dobriyan wrote:
> commit 5096add84b9e96e2e0a9c72675c442fe5433388a
> "proc: maps protection"
> 
> commit 831830b5a2b5d413407adf380ef62fe17d6fcbf2
> "restrict reading from /proc/<pid>/maps to those who share ->mm or can ptrace pid"
> 
> After Al added mm_for_maps(), maps_protect stopped controlling anything,
> because they're run at ->show time, but mm_for_maps() checks are done at
> ->start time.
> 
> Unless anyone objects, I'll remove maps_protect.

As long as this provides the same protections as maps_protect, I'm fine
with it.  I am a bit confused, though, since the reason I had to create
the sysctl entry in the first place was because akpm objected to the
maps file disappearing without a tunable.  Has this objection gone away?

> Also, logic behind second commit applies to /proc/*/pagemap , don't you
> think?

If that file shows memory location, yes.  What about the numa maps that
the first commit protects?

Thanks,

-Kees

-- 
Kees Cook                                            @outflux.net