On Sat, 13 Aug 2016, Patrick Schleizer wrote:

> "Off-Path TCP Exploits: Global Rate Limit Considered Dangerous"
> 
> https://regmedia.co.uk/2016/08/10/sec16_tcp_pure_offpath.pdf
> 
> https://security-tracker.debian.org/tracker/CVE-2016-5696
> 
> #####
> 
> Said to be fixed in linux 4.7, which is not yet available from any
> package sources.
> 
> The often quoted workaround for now:
> 
> /etc/sysctl.conf
> net.ipv4.tcp_challenge_ack_limit = 999999999
> sysctl -p
> 
> So it would also suffice if that fix was applied in sys-net?

Why do you think that sys-net in particular need this workaround?
Normally it is not an endpoint for any TCP connections? The workaround 
helps (and the attack works) only if it's applied to the kernel which
terminates the attacked TCP connection.

I don't think that it is such clear cut that sys-net is enough,
although attacking TCP connections that are terminated behind
NAT is somewhat more difficult. The attacker needs to have
a legimite connection to the VM behind NAT for challenge ack
counting purposes. To have such connection, the attacker would
need to trick the user to initiate a TCP connection to a host
under his control from the same VM as the attacked connection
(Fig 2 case from the paper). Whether that is realizable for the 
off-path attacker depends on TCP usage patterns of a VM.

> Should a Qubes security upgrade apply this workaround?


-- 
 i.

Reply via email to