[cisco-nas] >255 radius requests = bug?

[Dennis Peng]

achatz at forthnet.gr [achatz at forthnet.gr] wrote:
> The scenario
> -------------
> LNS terminating 500+ adsl users.
> The tunnel goes down/up, so all users are trying again to authenticate simultaneusly.
> Radius server isn't able to handle all those requests, so some udp packets are dropped.
> Router has to retransmit all these requests that aren't replied.
> Since unique-id is only 8 bits, we can have 255 concurrent unique access-requests.
>  
> Router sends a access-request using an id and at the same time the radius is using the same id 
> in order to reply to the router for a previous request (which also had this id). 
> So the router thinks that this reply from the radius is about the last request, 
> but this is actually for the previous request (both had the same  id).

This cannot happen. Each Access-Request the router sends has a random
number in the Request-Authenticator. When the RADIUS server responds,
the Message-Authenticator contains a hash of the RADIUS reply along
with Request-Authenticator and other fields. If the router has more
then one transaction outstanding for a particular transaction id, it
will try to decrypt the response packet with each of the outstanding
Request-Authenticator's which used that transaction id. Only the
"correct" Access-Request would have the right Request-Authenticator to
decrypt the packet. So you may see "decrypt" failures in the debug
logs, however, eventually we will match to the right one.

> The result
> ----------
> A user which is not allowed to login, will be authenticated normally and 
> will get all radius attributes of another user (who is allowed to login)!!!

The problem you are seeing may be a result of low memory. Your logs
show a failure to insert attributes in the RADIUS database, which is
usually caused by a lack of free memory. Have you looked at "show mem"
after one of these events to see if the amount of free memory was low?

> Can the above result be considered a bug from router's side?
> Is this the way radius authentication is supposed to work?

Because the transaction id number space is so limited (256 as you
mention), we extended our scalability by the use of multiple source
ports in later code so that bulk transactions with the RADIUS server
work better. Depending on what version of code you are running, this
may be on by default. In most recent code, we try to maintain backward
compatibility, so you have to configure the command "radius-server
source-ports extended". We will use source ports in the range from
21645 to 21844 in that case.

Dennis

> If yes, how can something like this be considered secure?

> _______________________________________________
> cisco-nas mailing list
> cisco-nas at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nas