[LibreQoS] Tracking unknown IPs (maybe for 1.4?)

Wed Nov 9 11:20:24 EST 2022

We have a bespoke solution to do similar (I keep meaning to make it more
generic and open source it). The basic operation is (using our Mimosa
devices as an example; it's actually a lot more complicated than that since
we have everything from apartment complexes with Ethernet jacks to regular
Ubiquiti devices in bridge mode):

   - A server contains an instance of FreeRADIUS.
   - Periodically, a script runs and queries UISP. It finds client sites
   with a device matching the type "Mimosa C5x" and an "other device" entitled
   "Service IP".
      - A radius record is then added for the CPE (using the MAC and IP
      from the Mimosa device), placing the Mimosa CPE on the IP address in the
      "mimosa" record.
      - A second radius record is added that matches Option 82 headers to
      see that a request passed through the Mimosa. This will always
hand out the
      IP address from the "Service IP" record.
      - The radius database is refreshed with this information.
   - When a CPE comes online, the DHCP server sends a RADIUS request. If
   the MAC address matches a RADIUS record, the device is assigned to the CPE
   address from the RADIUS records.
   - When a customer's device sends a DHCP request, it passes through the
   CPE. The CPE's "option 82" support decorates the DHCP request with the
   CPE's MAC address in a header. This then matches the second rule in Radius,
   ensuring that no matter what device the customer plugged in - it gets the
   Service IP.

This then dovetails into the QoS - because we can be 100% sure that the
customer's router has the IP address of the Service IP record in their
client site.

It took about an afternoon to setup, and is really nice. We have additional
rules like "place unknown CPEs into a block that is redirected to a page
reminding the installer to call in the account for setup", special handling
of suspended accounts and similar.

People have been *begging* Ubiquiti to a) support option 82 properly on the
M5/M2 line (I have a 10 year old request still unanswered!), and b) provide
some sort of RADIUS setup baked into UISP. The latter won't happen, because
it reduces vendor lock-in. But it's really easy to setup, and use UISP as a
"source of truth". (Obviously, when your clients are bridged you need to
take precautions - client isolation, switch port isolation and DHCP
snooping)

On Wed, Nov 9, 2022 at 10:07 AM dan <dandenson at gmail.com> wrote:

> How are you linking UISP to RADIUS?
>
> On Sat, Nov 5, 2022 at 10:29 AM Robert Chacón via LibreQoS <
> libreqos at lists.bufferbloat.net> wrote:
>
>> In our particular case we use RADIUS tied to UISP so we don't have the
>> immediate need, but I think it's an important feature to add.
>>
>> Perhaps cpumap-pping can have a feature to define "shaped subnets" during
>> the filter setup, and then we could query cpumap-pping for a JSON output of
>> IPs detected in traffic that are in the "shaped subnets" groups, but not
>> defined in the hash map.
>>
>> Curious to hear what others think here. Would others need this in order
>> to adopt LibreQoS?
>>
>>
>> On Sat, Nov 5, 2022 at 7:33 AM Herbert Wolverson via LibreQoS <
>> libreqos at lists.bufferbloat.net> wrote:
>>
>>> As we approach the v1.3 pre-release feature freeze, I've been thinking a
>>> little bit about nice things to have. One thing I found useful in both
>>> BracketQoS and Preseem was the ability to grab a list of IP addresses that
>>> had been through the shaper, but weren't mapped to a queue (obviously, only
>>> from within the "allowed IP" range - we're not trying to map the Internet!).
>>>
>>> In Preseem, there's a link to download a CSV file containing all the
>>> unmapped IP addresses and how much traffic they have consumed. BracketQoS
>>> (pre cpumap-pping) has a report showing the IPs (no traffic).
>>>
>>> *Why is this useful?*
>>>
>>> Knowing which local IP addresses were processed but not mapped lets you
>>> find:
>>>
>>> * the times that a device was installed, but the on-boarding process
>>> wasn't completed. Yes, that shouldn't happen. And - unfortunately - it
>>> occasionally does. If you're using RADIUS-based authentication, it's really
>>> difficult for this to happen - but not everyone is.
>>> * If there's a bug in your shaper integration, it's helpful to see
>>> "oops, I put X on the default"
>>> * Just occasionally, you get a customer who needs a special setup; it's
>>> helpful to see that it worked.
>>>
>>> *Current Status*
>>>
>>> Before cpumap-pping, Bracket was grabbing them by reading the pping
>>> output and listing addresses that didn't match a shaping rule. That doesn't
>>> work now:
>>>
>>> * xdp_pping is spitting out TC handles, rather than IP addresses.
>>> * With a default rule in place, and handling for IPv6 and IPv4 subnets,
>>> an IP address might not exactly match an entry (requires an LPM trie
>>> lookup) - and IPs matching a default rule (::/0 or 0.0.0.0/0) will
>>> always come back with the "default" handle.
>>>
>>> It's currently pretty tricky to do.
>>>
>>> So I'm curious; would others like to see this? I have a few ideas for
>>> how to make it work, but don't want to start serious planning/design if I'm
>>> the only one who wants the feature.
>>> _______________________________________________
>>> LibreQoS mailing list
>>> LibreQoS at lists.bufferbloat.net
>>> https://lists.bufferbloat.net/listinfo/libreqos
>>>
>>
>>
>> --
>> Robert Chacón
>> CEO | JackRabbit Wireless LLC <http://jackrabbitwireless.com>
>> Dev | LibreQoS.io
>>
>> _______________________________________________
>> LibreQoS mailing list
>> LibreQoS at lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/libreqos
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/libreqos/attachments/20221109/70a1cee6/attachment.html>