Hoi,

I know this is a long shot, but this afternoon I upgraded one of my routers at AS50869 from VPP 24.10 to VPP 25.10.0 (with the LinuxCP fix). Shortly there-after, two internet exchanges (both with a /23 peeringlan) complained that the router was flooding ARP requests. I could not see these in Linux CP, but they were visible in the Internet Exchange when looking at a packet dump. I could however see the ARP replies from the folks my router was flooding, like so:

root@nlams0:/etc/bird/ebgp/groups# time tcpdump -evni speedix arp
16:08:46.638960 44:4c:a8:c7:4a:33 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.223.39 is-at 44:4c:a8:c7:4a:33, length 46 16:08:46.881844 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:46.886507 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:46.902967 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:46.905873 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:46.940812 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:46.973942 44:4c:a8:c7:4a:33 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.223.39 is-at 44:4c:a8:c7:4a:33, length 46 16:08:46.983844 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:47.003539 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:47.010575 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:47.030974 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:47.069999 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:47.123048 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:47.134061 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:47.158991 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46 16:08:47.159000 c4:ca:2b:69:c8:f7 > b8:59:9f:e2:0a:9f, ethertype ARP (0x0806), length 60: Ethernet (len 6), IPv4 (len 4), Reply 185.1.222.21 is-at c4:ca:2b:69:c8:f7, length 46

In this network, the VPP router is b8:59:9f:e2:0a:9f and the router that was being flooded is at c4:ca:2b:69:c8:f7. They saw the traffic also. It immediately stopped when I set a static neighbor entry in Linux. I took a trace from dpdk-input but it did not reveal any outbound ARP traffic (which makes sense). It did however show the ARP replies.

Could it be that between 24.10 and 25.10 release, something changed in the ARP handling that might trigger an ARP flood from within arp request/reply/ip-neighbor code? I'm hoping somebody can remember any changes, I scanned over a bunch of changes but a year is a long time and bisecting on an internet exchange is impractical. I may be able to repro this behavior in a lab, but before I go deeper: does this ARP flooding ring a bell for anybody ?

groet,
Pim

--
Pim van Pelt <[email protected]>
PBVP1-RIPE https://ipng.ch/

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#26582): https://lists.fd.io/g/vpp-dev/message/26582
Mute This Topic: https://lists.fd.io/mt/116519525/21656
Group Owner: [email protected]
Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/14379924/21656/631435203/xyzzy 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to