Andy Green wrote:
Christopher Aillon wrote:
> The kernel has more vulnerabilities than this user-space
> application does. Let's reconsider having that in the distro, too.
With respect this is not a good response to my question. How many
kernel problems are remote-exploitable? Does the kernel of itself visit
random external "scripts" on the Internet and execute what it finds
there? No. But a browser is designed to do such actions. If we really
do talk about code of such complexity that "MASSIVE changes which took
several architects months to perfect, and it STILL caused 10-20
regressions" it's a lot more frightening to hear that about usermode
code that exists to go out to a potentially hostile Intenet on behalf of
a logged-in user than it is to hear the same about a kernel where the
vast bulk of vulns are local only. Objectively, looking at your
description of security fixes on the beast, shouldn't people take pause
at a creature that is so complex and poorly understood, but is our main
proposed way of interfacing to the good and evil of the external world?
You're grossly overreacting. If you re-write the way the browser
surface, I'd say you did a pretty damn good job. The issue and code is
extremely well understood by those that need to understand it, myself
included as I discovered and provided a workaround for this issue while
I interned at Netscape. As an intern, I just didn't have the time to
fix it properly, and still don't have time to backport it if I'm
expected to do all I do as it is.
There's always a potential for security problems when you don't control
the input you get.