Open ikarus23 opened 9 years ago
Do you know if there's a way to fix this?
Until now I only know of some iptables
magic or fail2ban
configs (see this post). I can look into this more this weekend.
Sorry for not updating this earlier. I did some additional search on that weekend, but with no remarkable realizings. It seams that every application has to "fix" this on its own. The fix is to deactivate client-initiated SSL/TLS session renegotiations. Some software like Apache or NGINX have a setting to do this, others don't. Other argue that "it is the responsibility of server deployments, not a security library, to prevent or limit renegotiation when it is inappropriate within a specific environment." Red Hat rated this issue with a low security risk (link). This blog has a nice article including prevention and mitigation measures. But they are all not really "satisfying"...
The way I see it there are only tow options:
iptables
rule to mitigate the risk.Ok, did some more real word testing to see how bad this is. By using the THC tool (patched to work with STARTTLS) I was able to increase the the CPU usage of my vServer. After some time however, it averaged out around 40%. All my services were still available during this test.
So maybe it's not so bad after all... It would be nice if other members of the mail in a box community could test this too. If they come to the same result, I would suggest to ignore this issue.
Multiple configurations allow client-initiated SSL/TLS session renegotiations. This leads to a denail of service vulnerability (CVE-2011-1473) that is really easy to exploit with tools like THC SSL DoS (and this patch).
Affected services:
This was validated using OpenSSL (and the TCH SSL DoS tool).
box.occams.info:587 tested with
sslyze
(example; the other ports work as well):(The web server (
nginx
) is not affected by this issue.)