FreeRADIUS server and the Simultaneous-Use parameter.
0. INTRODUCTION
Lots of people want to limit the number of times one user account can
login, usually to one. This is hard to do with the radius protocol;
the nature of the accounting stuff is such that the idea the radius server
has about the list of logged-in users might be different from the idea
the terminal server has about it.
However, most terminal servers have an alternative way to get a list
of logged-in users. Most support some way through telnet, some have
a finger-daemon builtin and a lot of them support SNMP. So if the
radius server thinks that someone is trying to login a second time,
it is possible to check on the terminal server itself if the first
login is indeed still active. Only then access is denied for the
second login.
1. PREREQUISITES
You need to have perl installed.
For SNMP checks, you have 2 options. You can use the `snmpget' program
from the cmu-snmp tools. You can probably get precompiled ones,
maybe even packaged for your system (Debian/Linux, Redhat/Linux, FreeBSD
ports collection etc). The source code is at
http://www.net.cmu.edu/projects/snmp/snmpapps/. The Linux-specific
version of this is at http://www.gaertner.de/snmp/
The other option is to install the SNMP_Session and BER modules that
for example the well known `mrtg' package uses. This is recommended.
In that case you need no external snmpget program, checkrad will
speak SNMP directly. See http://www.switch.ch/misc/leinen/snmp/perl/
The checkroutine for USR/3Com Total Control racks uses the Net::Telnet
module from CPAN, at least version 3.00. If you need that, obtain it from
your local CPAN mirror (or see http://www.perl.com/CPAN/). The checkrad.pl
perl script will autodetect if that module is installed.
2. USAGE.
It works by adding the `check' parameter "Simultaneous-Use" to the entry
for a users or DEFAULT in /etc/raddb/users. It should be at least one;
it defines the maximum number of users logged in with the same account name.
For example:
#
# Simultaneous use restrictions.
#
DEFAULT Group == "staff", Simultaneous-Use := 4
Fall-Through = 1
DEFAULT Group == "business", Simultaneous-Use := 2
Fall-Through = 1
DEFAULT Simultaneous-Use := 1
Fall-Through = 1
NOTE!!! The "Simultaneous-Use" parameter is in the "check" A/V pairs,
and not in the Reply A/V pairs (it _is_ a check).
For SQL, after creating and populating your schema, you should
execute the following statement (for MySQL, others may vary):
INSERT INTO radgroupcheck (GroupName, Attribute, op, Value) values("dialup", "Simultaneous-Use", ":=", "1");
Once that is done, your users should be limited to only one login at a time.
3. IMPLEMENTATION
The server keeps a list of logged-in users in the /var/log/radutmp file.
This is also called "the session database". When you execute "radwho",
all that radwho really does is list the entries in this file in a pretty
format. Only when someone tries to login who _already_ has an active
session according to the radutmp file, the server executes the perl
script /usr/local/sbin/checkrad (or /usr/sbin/checkrad, it checks for
the presence of both and in that order). This script queries the terminal
server to see if the user indeed already has an active session.
The script uses SNMP for Livingston Portmasters and Ciscos, finger for
Portslave, Computone and Ascend, and Net::Telnet for USR/3Com TC.
Since the script has been witten in perl, it's easy to adjust for
any type of terminal server. There are implementations in the script for
checks using SNMP, finger, and telnet, so it should be easy to add
your own check routine if your terminal server is not supported yet.
You can find the script in the file src/checkrad.pl.
You need to set the correct type in the file /etc/raddb/naslist so that
checkrad KNOWS how it should interrogate the terminal server. At this
time you can define the following types:
type Vendor Uses method needs Need naspasswd
==== ====== =========== ===== ==============
ascend Lucent SNMP SNMP No
bay Nortel finger finger command No
cisco Cisco SNMP SNMP Optional [1]
computone Computone finger finger command No
cvx Nortel SNMP SNMP No
digitro Digitro rusers rusers command No
livingston Livingston SNMP SNMP No [2]
max40xx Lucent finger finger command No
netserver USR/3com telnet CPAN Net::Telnet Yes
pathras Cyclades telnet CPAN Net::Telnet Yes
patton Patton SNMP SNMP No
portslave ? finger finger command No
pr3000 Cyclades SNMP snmpwalk command No
pr4000 Cyclades SNMP snmpwalk command No
tc USR/3com telnet CPAN Net::Telnet Yes
usrhyper USR/3com SNMP SNMP No [3]
versanet VersaNet SNMP SNMP No
other none N/A - No
[1] In naspasswd file: set username to SNMP, password is community.
[2] Needs at least ComOS 3.5, SNMP enabled.
[3] Set "Reported Port Density" to 256 (default)
"other" means "don't bother checking, I believe what radutmp says".
This really is not recommended, if a user has a "stuck" entry in the
session database she will not be able to login again - hence the
extra check that "checkrad" does.
4. IF IT DOESN'T WORK
Note that you need to add the Simultaneous-Use parameter to the
check item (first line), not the reply item, using the ':=' operator.
You can edit the `checkrad' perl script and turn on debugging. Then
watch the debug file. The `radius.log' file also gives some hints.
You can also run the "checkrad" script manually, use the "-d"
switch to get debug output on standard output instead of in the log.
See also:
http://wrath.geoweb.ge/simult.html
which has a good discussion of the use of Simultaneous-Use.
5. CAVEATS
This solution checks the radutmp file. This file is kept up-to-date from
the Accounting records the NAS sends. Since some NASes delay these records
for quite some time, it is possible to get a double login by logging in
twice at _exactly_ the same time (plus or minus the mentioned delay time),
since neither of the logins are registered yet.
The solution would be to create a small 1-minute cache of Authentication
records, that is also checked for double login attempts. Perhaps in the
next version.
When implementing this one thing was considered the most important: when
trying to detect double logins, we always try to err on the safe side. So
in rare cases, a double login is possible but we try never to limit access
for a legitimate login.
6. PROBLEMS WITH DROPPED CONNECTIONS
Our PM3, with 2 ISDN-30 lines coming into it, had the habit of sometimes
dropping connections. In a few cases, the portmaster thought the session was
still alive so if the user tried to login again, he or she was denied access.
In our case, this problem was caused by a bad PRI line from the phone
company.
We tried to compensate this by setting the Idle-Timeout to 15 minutes. That
way, even if a user did get locked out the portmaster would clear the rogue
session within 15 minutes and the user could login again.