Re: [gentoo-dev] Re: Enemy Territory and Gentoo

[Martin Schlemmer]

On Mon, 2003-09-22 at 19:44, Paul de Vrieze wrote:

> If there were some way that we can have overlay filesystems supported by the
> kernel ( readonly mount root, and over that mount a freshly created dir that
> will be used for all writes instead of the original. For reads though the
> original filesystem is visible (as long as there is no file with the same
> name in the writable part)) then it would be quite easy with chroot to
> "track" changes. I'm not a kernel coder, and I have no idea whether such code
> allready exists. It should be not too complex either. And also provide some
> of the "extra security" that was asked by an earlier thread this month. It
> should also be more foolproof than the sandbox, but relies on the kernel.
>

I have thought about using an kernel module multiple times (there are
actually one or two of those apps that tracks installs that use such
an module ... or tried to develop one back than).  Problem is though
that you will either limit the user to what kernel he use, or might
run into issues with having to update it all the time to not break on
new kernels or way different trees like -aa, etc.  Do not know how
much these hold anymore though.

The bigger issue, is that late 2.5 and now 2.6 kernels do not allow
you to replace system calls (something needed for a module like this).
This will then either force us to not use this for 2.6, or hack the
kernel, which will really limit the user.  On another note - I do not
know if the new API and hooks added for the security modules might
enable us to have the same end result as hooking system calls with
our own would have done ...


Regards,

-- 

Martin Schlemmer
Gentoo Linux Developer, Desktop/System Team Developer
Cape Town, South Africa