mailing list archives
Re: Apple Safari ... DoS Vulnerability
From: Michal Zalewski <lcamtuf () coredump cx>
Date: Thu, 26 Feb 2009 19:10:09 +0100
The fun times of security semantics!
Old debates never die...
Vulnerabilities are a subset of software engineering bugs. As the name
implies, they are defined strictly by the impact they have; if a bug
does not render the victim appreciably susceptible to anything that
would be of value to external attackers, it is not a security problem.
Now, there are two points to be made:
1) "Value to the attacker" is a broad and fuzzy term that also covers
emotional gratification (by just causing hardship to a disliked party)
- so loss of availability should be often treated as a security glitch
(well, you could also say a "reliability glitch" and start another
argument); but the important thing is, not all bugs that cause a crash
will cause noticeable loss of availability - i.e., no service is
denied or deferred to third parties. For example, crashing a sshd or
ftpd child handling my own connection is not interesting by itself,
unless events leading to the crash, or the crash itself, impose a
significant and repeatable resource strain. Crashing a keep-alive
httpd child might be marginally more expensive, and hence maybe a
limited security concern.
2) "Appreciably susceptible" is just as hard to quantify when dealing
with high loss, but low probability scenarios; there were quite a few
bugs that likely affected very few or no users (e.g., many of the
publicly reported command-line overflows in non-suid programs), but a
hypothetical scenario where it would matter could be constructed (in
the aforementioned case, say, really bad PHP / CGI scripting). Most
people dismiss such vulnerability reports, but it's difficult to draw
Anyway... bottom line is, any attempts to formalize the criteria are
bound to fail (and have mostly failed in the past), and common sense
is the best tool we have.
Full-Disclosure - We believe in it.
Hosted and sponsored by Secunia - http://secunia.com/