[Gcl-devel] Re: GCL packaging

[Camm Maguire]

Greetings!

Gérard Milmeister <address@hidden> writes:

G> Hi Camm,
> 
> I am the Fedora package maintainer of GCL. I have
> trouble building it on anything other than i386.
> Fedora supports i386, x86_64, ppc and ppc64. It
> used to work on x86_64, but lately it doesn't anymore
> there either.
> We have a patch (or rather a hack) to change
> personality, that used to work.
> Anyway, I looked at the Debian patch, which is,
> unfortunately, enormous, and I cannot figure
> out the bits that are relevant (i.e., not Debian
> specific).

This is because the Debian 2.6.7 package is in fact a testbed for
version 2.6.8pre.

> What are the bits that need to be adapted for
> x86_64 and the PowerPC platforms.

I'm not sure what failure you are reporting here.  More details,
please.  

The critical issue is that gcl needs a non-randomized traditional sbrk
function.  GCL attempts to automatically detect and disable this with
a personality change and exec at the top of main.  This has worked on
fedora for quite a while now.

GCL does not execute code on the C stack, but does execute code in an
sbrk-expanded .data section.  This needs to be allowed in selinux and
the like, but will probably result in log messages, AFAICR.

Unfamiliar with textrel_shlib_t -- would appreciate education.  Also
would appreciate knowing if and when the strategies being implemented
by the various kernels to protect binaries from themselves become
stabilized, and if so on what model. 

Take care,


> Also, there is SELinux to be taken care of (we had
> to change the context of the saved_ansi_gcl binary
> to textrel_shlib_t) and the various security bits
> available in Fedora, like execstack. I do not know
> how much these apply to Debian.
> 
> Regards,
> Gérard Milmeister
> 
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah