option FIB_ALGO and dpdk_lpm4

[Marek Zarychta]

W dniu 08.02.2021 o 19:32, Alexander V. Chernikov pisze:
> 08.02.2021, 14:33, "Marek Zarychta" <zarychtam at plan-b.pwste.edu.pl>:
>> W dniu 08.02.2021 o 13:10, mike tancsa pisze:
>>>  I have been setting up some tests to see if
>>>
>>>  option FIB_ALGO and dpdk_lpm4.ko
>>>
>>>  will help with my pkt forwarding needs and large routing tables. So far so good. But one thing I noticed, is that its very chatty to dmesg.
>>>  eg
>>>  alloc_nhgrp: new mpath group: num_nhops: 2
>>>  compile_nhgrp: O: 2/2
>>>  compile_nhgrp: OO[0]: 1/1 curr=1 slot_idx=0
>>>  compile_nhgrp: OO[1]: 0/0 curr=1 slot_idx=1
>>>  alloc_nhgrp: new mpath group: num_nhops: 2
>>>  compile_nhgrp: O: 2/2
>>>  compile_nhgrp: OO[0]: 1/1 curr=1 slot_idx=0
>>>  compile_nhgrp: OO[1]: 0/0 curr=1 slot_idx=1
>>>  alloc_nhgrp: new mpath group: num_nhops: 2
>>>  compile_nhgrp: O: 2/2
>>>  compile_nhgrp: OO[0]: 1/1 curr=1 slot_idx=0
>>>  compile_nhgrp: OO[1]: 0/0 curr=1 slot_idx=1
>>>
>>>  are these debugging messages that forgot to be turned off ? What do they mean ?
>>>  Thanks for this work!
>>>
>>>  13.0-STABLE #11 stable/13-cc1352c1f-dirty
>>
>> Thank you for sharing this Mike. Could you please reveal us how do you
>> feed your routing tables? Is net/bird{,2} or net/frr7 involved? Any
>> problems or hints to make the routing daemon working with new routing stack?
> Non-multipath should work as before, multipath works for quagga/frr but needs some patches for bird.

Thank you for the clarification, so is with anything but quagga or frr
the sysctl setting net.route.multipath=0 obligatory now?
>>
>> The new routing stack looks very promising, please let me also give this
>> way some appreciations to melifaro@ and other people who worked on it.
>>
>> I was also trying to test it with legacy net/bird and multiple fib
>> tables, but I was early hit by: "KRT: Error sending route x.x.x.x/y to
>> kernel: Operation not supported"
> Any chance you could clarify what are these routes? "Operation not supported" looks a bit weird, it shouln't happen.
>> Setting net.add_net.add_addr_allfibs=1addr_allfibs=1 changed it a bit,
>> but still some blackhole /32 routes seem to get rejected.
> Just "blackhole" route in the bird config? /32 or all?

I used for tests the feed from Peter Hessler's OpenBSD spam trapping
project[1]. On FreeBSD 11.4 I see these routes in net/bird as
blackholed, for example:
x.x.x.x/32  blackhole [bgp_spamd 20:20:43 from y.y.y.y] * (100) [ASzzzz]
They work the same was as routes added by route(8) with option "-blackhole"

With new routing stack, these routes are rejected with the message
above. Now in net/bird, they appear like the example below and import to
the fib (fib number is not equal to 0 in this case) is blocked:
x.x.x.x/32 unreachable [SPAM 19:58:18 from y.y.y.y] ! (100/-) [ASzzzz]

Probably it all should be tested in normal peering, but my initial test
was performed on the old lab setup where multiple fibs and policy
routing[2] were involved. The config was loosely based on the example by
Ondrej Filip from the[2].

Once again thank you for implementing all these improvements into
FreeBSD routing stack and please don't get me wrong, I am just testing
it a bit before migration from 11.4-STABLE, but not complaining about
anything.

[1] http://rs.bgp-spamd.net/client/index.html
[2] https://gitlab.nic.cz/labs/bird/-/wikis/Policy_routing

-- 
Marek Zarychta

-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 495 bytes
Desc: OpenPGP digital signature
URL: <http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20210208/1f14d8a4/attachment.sig>