Backport fix for BZ #18928?

[Florian Weimer]

On 12/17/2015 05:54 PM, Carlos O'Donell wrote:

> I would be interested in hearing Florian and Joseph's opinoins on my suggestion
> below since this is a topic that relates to our rolling release branches.

If we can switch Fedora to rebase to point releases, rather than
maintaining our own set of patches, that would be nice.  Otherwise, I
have no particularly preference.

> In the glibc-2.5 era the LD_POINTER_GUARD variable was actually a workaround
> for broken Sybase binaries that could not handle the encrypted pointers. I assume
> they inspected libraries internals for reasons that were never clearly explained.
> That should not be the case anymore, but we might break such programs in 2.22
> if we remove LD_POINTER_GUARD after the release and that's a compatibility issue
> for the stable release.

Hector pointed out that on i386 and x86-64, LD_POINTER_GUARD does not
actually disable pointer mangling completely:

  <https://sourceware.org/ml/libc-alpha/2015-10/msg00425.html>

So you will not have much luck running those old Sybase binaries, with
or without LD_POINTER_GUARD.

> The up side is that this is a security issue, and that no conforming application
> should be able to see a difference.

If you treat this as a security issue, you also want the tls_dtor_list
mangling (commit f586e1328681b400078c995a0bb6ad301ef73549 on master).

Florian