accounting records per nasid, not ip address
hi, this is my first time using freeradius. had a good experience so far. i had to coerce the latest cvs version into fedora core 3 with more hacks than i would have liked, but that's ok, it's up and running. one issue i find so far is that my NASs have a dynamic ip. i managed this externally at great pains, however, for the purposes of radius accounting, i'd like to see the accounting records on a per-nasid file in /var/log/radius/radacct/<nasid> rather than what i see now as: /var/log/radius/radacct/<ip-address> i could not find a simple way to do this by looking around, though i may be missing something. is this doable as-is? if not, any hint on where to hack this would be greatly appreciated. while i am on the list, and at the risk of irking someone i would like to (lazily) ask if anyone knows offhand where can i find a document that i saw a few months back on "hardening" radiusd. the process exited on me a couple of times due to dynamic ip issues, creating multi-hour blackouts that i rather not see again. looking for freeradius hardening or crashing i cannot seem to find that most excellent list of things to restart or manage a crashed server. thanks in advance!
freeradius@puchol.com wrote:
i had to coerce the latest cvs version into fedora core 3 with more hacks than i would have liked,
Such as...? It's a development version, so not everything works.
one issue i find so far is that my NASs have a dynamic ip.
That's evil. Very evil.
i managed this externally at great pains, however, for the purposes of radius accounting, i'd like to see the accounting records on a per-nasid file in
/var/log/radius/radacct/<nasid>
rather than what i see now as:
/var/log/radius/radacct/<ip-address>
Step 1: Find out which module writes those files Step 2: Edit the modules configuration See "radiusd.conf". You're probably looking for th e"detail" module.
while i am on the list, and at the risk of irking someone i would like to (lazily) ask if anyone knows offhand where can i find a document that i saw a few months back on "hardening" radiusd. the process exited on me a couple of times due to dynamic ip issues, creating multi-hour blackouts that i rather not see again. looking for freeradius hardening or crashing i cannot seem to find that most excellent list of things to restart or manage a crashed server.
doc/supervise-radiusd.txt Alan DeKok.
We are using freeradius 1.0.4 and having a problem with VSA id's higher than 255 (more than 8 bit). ATTRIBUTE Ascend-Http-Redirect-URL 287 string Lucent ATTRIBUTE Ascend-Http-Redirect-Port 288 integer Lucent When one of the attributes above are sent back to the NAS, it seems that radius is sending it as 8 bit thus interpreted as something else on the Lucent NAS. Any simple solution for that ? Best Regards / Fawaz
Fawaz Qamhawi <fawazq@eim.ae> wrote:
We are using freeradius 1.0.4 and having a problem with VSA id's higher than 255 (more than 8 bit).
Hmm.. those aren't in the standard dictionaries that I can see.
When one of the attributes above are sent back to the NAS, it seems that radius is sending it as 8 bit thus interpreted as something else on the Lucent NAS.
The code in src/lib/radius.c is responsible. It should be checking for attributes above 255, and discarding them.
Any simple solution for that ?
How are the attributes supposed to be encoded in the packet? The normal VSA's use one byte to represent vendor attributes. Since 287 won't fit into ne byte, something else has to be done here. Alan DeKok.
Ascend (as Lucent) has been introducing tags with values higher than 256 in the VSA's for a while (first message I saw where the problem of long tags was mentioned was from January 2004). An example from their dictionary shows: ATTRIBUTE Ascend-MOH-Timeout 261 integer The format for the long tag VSA is the same as the standard Vendor-Specific attribute (8 bit tag, 8 bit length) but the sub-attribute tag field has been expanded to 16 bits. The sub-attribute length field remains 8 bits. All vendor specific attributes are coded using 16-bit attribute type in network byte order and Lucent-Vendor-Id (4846) as Vendor-Id. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Attr Type | Length | Vendor-Id +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Vendor-Id (cont) | Vendor Type(16-bit) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Vendor Length | Vendor-value...... +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+- I believe the support for long Vendor-Specific tags was discussed here in the past with limited interest in support. It seems that this is on a NAS by NAS basis and only some of the VSA's are using the 16 bit tags. The solution seems to be to indicate that long tags are used by this NAS for particular vendors. Something like: 192.168.1.1 ... VendorLongTags=Ascend - indicating that Ascend VSA's use long tags and all other VSA's like Cisco) would be short. Ascend / Lucent VSA's do not always use long tag VSAs. This introduction of long tags is a real wart for every RADIUS server. There are probably other ways to have avoided 16 bit tags. Naturally the offender is too big to ignore and arbitrarily forced the issue. Remember that in the past Ascend (pre-Lucent) grabbed unassigned RADIUS attributes (from 119 to 255) without thinking there might be a problem with that either. Alan DeKok wrote:
Fawaz Qamhawi <fawazq@eim.ae> wrote:
Any simple solution for that ?
How are the attributes supposed to be encoded in the packet? The normal VSA's use one byte to represent vendor attributes. Since 287 won't fit into ne byte, something else has to be done here.
Alan DeKok.
- List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html
Michael Lecuyer <mjl@theorem.com> wrote:
The format for the long tag VSA is the same as the standard Vendor-Specific attribute (8 bit tag, 8 bit length) but the sub-attribute tag field has been expanded to 16 bits. The sub-attribute length field remains 8 bits.
That doesn't sound too bad.
All vendor specific attributes are coded using 16-bit attribute type in network byte order and Lucent-Vendor-Id (4846) as Vendor-Id.
That makes it easier.
I believe the support for long Vendor-Specific tags was discussed here in the past with limited interest in support.
It's about 40 lines of code to support. The weirdness that I recall was Nortel, which mixed normal VSA's, and USR-style VSA's in the same vendor space.
192.168.1.1 ... VendorLongTags=Ascend - indicating that Ascend VSA's use long tags and all other VSA's like Cisco) would be short. Ascend / Lucent VSA's do not always use long tag VSAs.
If it's always that the Lucent attributes use 16-bit id's, it's OK.
This introduction of long tags is a real wart for every RADIUS server. There are probably other ways to have avoided 16 bit tags. Naturally the offender is too big to ignore and arbitrarily forced the issue. Remember that in the past Ascend (pre-Lucent) grabbed unassigned RADIUS attributes (from 119 to 255) without thinking there might be a problem with that either.
Yup. I'll add something to the CVS head. Grab a snapshot in a few days, and see if it works. Alan DeKok.
participants (4)
-
Alan DeKok -
Fawaz Qamhawi -
freeradius＠puchol.com -
Michael Lecuyer