[VPP-1685] vpp ping dropping icmp reply

Open vvalderrv opened this issue 10 months ago • 4 comments

On latest master (I am testing 9febbcdc), ping from cli seems to ignore icmp replies which get dropped:

create veth & assign Linux host IP
~# sudo ip link add dev vm1 type veth peer name vm2

~# sudo ifconfig vm2 1.1.1.2/24 up
run VPP, disabling rdma to be sure
~# $PWD/build-root/install-vpp_debug-native/vpp/bin/vpp "unix { interactive } plugins

Unknown macro: { plugin rdma_plugin.so { disable } }

"
DBGvpp# create host-interface name vm1

DBGvpp# set int ip addr host-vm1 1.1.1.1/24

DBGvpp# set int st host-vm1 up

DBGvpp# clear trace

DBGvpp# trace add af-packet-input 10

DBGvpp# ping 1.1.1.2

Statistics: 5 sent, 0 received, 100% packet loss

DBGvpp# sh tr
- - - - Start of thread 0 vpp_main -------------------
        Packet 1
        
        </li> </li> </li> </li> </li> </li> </li> </li> </li> </li> </li> </li> </li> </li> </li> </li> </li> </li>

00:00:01:043605: af-packet-input

af_packet: hw_if_index 1 next-index 4

tpacket2_hdr:

  status 0x20000001 len 42 snaplen 42 mac 66 net 80

  sec 0x5ce6a84c nsec 0x28be2a1f vlan 0 vlan_tpid 0

00:00:01:044117: ethernet-input

ARP: f2:dc:63:9b:18:c2 -> 02:fe:61:ef:48:39

00:00:01:044171: arp-input

reply, type ethernet/IP4, address size 6/4

f2:dc:63:9b:18:c2/1.1.1.2 -> 02:fe:61:ef:48:39/1.1.1.1

00:00:01:044969: error-drop

rx:host-vm1

00:00:01:044980: drop

arp-input: ARP request IP4 source address learned

Packet 2

00:00:02:050854: af-packet-input

af_packet: hw_if_index 1 next-index 4

tpacket2_hdr:

  status 0x20000001 len 110 snaplen 110 mac 66 net 80

  sec 0x5ce6a84d nsec 0x294bbace vlan 0 vlan_tpid 0

00:00:02:051362: ethernet-input

IP4: f2:dc:63:9b:18:c2 -> 02:fe:61:ef:48:39

00:00:02:051380: ip4-input

ICMP: 1.1.1.2 -> 1.1.1.1

tos 0x00, ttl 64, length 96, checksum 0xd847

fragment id 0x9e51

ICMP echo_reply checksum 0x8665

00:00:02:051415: ip4-lookup

fib 0 dpo-idx 5 flow hash: 0x00000000

ICMP: 1.1.1.2 -> 1.1.1.1

tos 0x00, ttl 64, length 96, checksum 0xd847

fragment id 0x9e51

ICMP echo_reply checksum 0x8665

00:00:02:051430: ip4-local

ICMP: 1.1.1.2 -> 1.1.1.1

  tos 0x00, ttl 64, length 96, checksum 0xd847

  fragment id 0x9e51

ICMP echo_reply checksum 0x8665

00:00:02:051437: ip4-icmp-input

ICMP: 1.1.1.2 -> 1.1.1.1

tos 0x00, ttl 64, length 96, checksum 0xd847

fragment id 0x9e51

ICMP echo_reply checksum 0x8665

00:00:02:051444: vpe-icmp4-oam

OAM: rx from address 1.1.1.2, target index -1

00:00:02:051451: error-drop

rx:host-vm1

00:00:02:051458: drop

vpe-icmp4-oam: icmp4 replies dropped (no registration)

Benoît Ganne

Benoît Ganne

bganne (Thu, 23 May 2019 19:42:08 +0000): The order of initialization changed recently and oam got initialized after ping, overwriting next node with its own for icmp reply. So oam was broken for a long time but nobody noticed.
nranns (Thu, 23 May 2019 17:20:52 +0000): "vpe-icmp4-oam" how did that get there?
bganne (Thu, 23 May 2019 14:26:40 +0000): This was initially reported by Thomas on real HW with rdma driver: https://gist.github.com/tfherbert/c5b1ece10b1d37ca9df0752925f7f53c

Original issue: https://jira.fd.io/browse/VPP-1685

Feb 02 '25 04:02 vvalderrv