mailing list archives
Re: Linux Distribution Recomendation
From: Vincent <pros-n-cons () bak rr com>
Date: Fri, 12 Mar 2004 18:44:01 -0800
On Thu, 11 Mar 2004 10:47:41 +0100
peter () devbox adamantix org (Peter Busser) wrote:
I probably shouldn't reply to this as its trivial & getting offtopic but a few things stick out.
[About security features costing a significant amount of performance]
Fair enough, significant was the wrong word, I was trying to recall what I
learned between you and Ingo from the debian-devel list a few months back.
Now I see it was the VM issue and full compatibility that still had hurdles.
Yeah, this stuff has more impact on the compatibility level than on the
performance level. But it is possible to make even the Sun Java environment
to work on a PaX kernel with a bit of tweaking, so it isn't all that bad.
Especially because most programs and libraries simply work without any extra
The debian-devel discussion was mostly about Russel Coker and Ingo's claims
that his patch does everything that PaX does, without breaking compatibility.
That is simply not true. It provides less protection and even then still
breaks compatibility. You cannot download the XFree86 source code, recompile
it with ELFLoader module support and run it as is on his kernel patch. I
respect the fact that people make trade-offs, OpenBSD made similar trade-offs.
Basically they trade in a bit of security for a bit of compatibility. That is
ok, if compatibility is more important than security.
Are we talking about the same thread?
In this one Ingo explicitly states a few times PaX is more secure than exec-shield.
Like it doesn't support other arch's and will never attempt to fix lib bss data.
Also the XFree problem was a bug in XFree it relied on execution on non exec area.
With my admittingly limited knowledge of the subject I interpret this to be
something like the old standards compliant GCC issue where it broke apps, but
only cause those apps were relying on a broken implementation. He said X was
used properly on other archs but on i386 it did not. I guess cause read-only
memory can be executed people used it which sounds like a design flaw to me.
I use fedora and have heard people mention breakage of wine with exec-shield
enabled but haven't looked into why so it obviously isn't 100%. This is the
'middle ground' stuff you were talking about, not as good as others, but better
than the default. Now I'm off to actually get informed on this whole subject =)
What troubles me is the lack of openness about it. I mean, some people try to
make it look as if there is no trade-off, i.e. that they provide full security
AND full compatibility. That is simply not true.
So the point that security almost always asks for something in return holds true
to some degree.
Re: Linux Distribution Recomendation Michael Gale (Mar 04)
RE: Linux Distribution Recomendation Neil Fryer (Mar 04)
- RE: Linux Distribution Recomendation, (continued)