gpe network not work

[Junbo Jiang]

Hi,
Maybe this is not a bug, I have an lab with two nodes, one controller and one compute both running vpp and vpp-agent, and both configured
[ml2]
type_drivers = flat,vlan,gpe
[ml2_vpp]
physnets=physnet1:GigabitEthernet0/5/0
etcd_host=10.4.1.5
etcd_port=2379
gpe_locators=physnet1
gpe_src_cidr=10.4.3.5/24 # 10.4.3.6/24 for another node
enable_l3_ha=false

openstack network create --provider-network-type gpe --provider-segment 100 net1
openstack subnet create --network net1 --subnet-range 10.0.1.0/24 net1-subnet

nova boot --image centos --flavor m1.small --nic net-name=net1 --availability-zone nova:node1 test1
nova boot --image centos --flavor m1.small --nic net-name=net1 --availability-zone nova:node2 test2

after test1(10.0.1.56) and test2(10.0.1.137) start and I manually configured ip address on it, test1 failed to ping test2, there is no error in vpp, vpp-agent.

here is what I found now,
both test1 and test2 got the arp reply(it's look like the vpp send the arp reply directly)
[root@host-10-0-1-56 ~]# arp -n
Address HWtype HWaddress Flags Mask Iface
10.0.1.2 ether fa:16:3e:4a:85:a1 C eth0
10.0.1.1 ether de:ad:00:00:00:00 C eth0
10.0.1.137 ether fa:16:3c:d4:fe:0a C eth0

the 'trace add vhost-user-input' on node2 showed

00:02:31:927340: vhost-user-input
     VirtualEthernet0/0/0 queue 0
   virtio flags:
    INDIRECT Indirect descriptor
   virtio_net_hdr first_desc_len 12
     flags 0x00 gso_type 0
     num_buff 0
00:02:31:927355: ethernet-input
  frame: flags 0x1, hw-if-index 4, sw-if-index 4
  IP4: fa:16:3e:90:75:9c -> fa:16:3c:50:8f:40
00:02:31:927357: l2-input
  l2-input: sw_if_index 4 dst fa:16:3c:50:8f:40 src fa:16:3e:90:75:9c
00:02:31:927360: l2-input-feat-arc
  IN-FEAT-ARC: head 1 feature_bitmap 140295 ethertype 800 sw_if_index 4, next_index 21
00:02:31:927362: acl-plugin-in-ip4-l2
  acl-plugin: lc_index: -1, sw_if_index 4, next index 1, action: 3, match: acl -1 rule 1 trace_bits 80000000
  pkt info 0000000000000000 0000000000000000 0000000000000000 3801000a8901000a 0004030100000008 0200ffff00000000
   lc_index 0 l3 ip4 10.0.1.137 -> 10.0.1.56 l4 lsb_of_sw_if_index 4 proto 1 l4_is_input 1 l4_slow_path 1 l4_flags 0x03 port 8 -> 0 tcp flags (invalid) 00 rsvd 0
00:02:31:927366: l2-input-feat-arc-end
  IN-FEAT-ARC: head 0 feature_bitmap 40295 ethertype 0 sw_if_index -1, next_index 16
00:02:31:927367: l2-input-acl
  INACL: sw_if_index 4, next_index 9, table 5, offset 1392
00:02:31:927369: l2-learn
  l2-learn: sw_if_index 4 dst fa:16:3c:50:8f:40 src fa:16:3e:90:75:9c bd_index 1
00:02:31:927372: l2-fwd
  l2-fwd: sw_if_index 4 dst fa:16:3c:50:8f:40 src fa:16:3e:90:75:9c bd_index 1 result [0xffffffffffffffff, -1] static age-not bvi filter learn-event learn-move
00:02:31:927373: l2-flood
  l2-flood: sw_if_index 4 dst fa:16:3c:50:8f:40 src fa:16:3e:90:75:9c bd_index 1
00:02:31:927374: l2-output
  l2-output: sw_if_index 3 dst fa:16:3c:50:8f:40 src fa:16:3e:90:75:9c data 08 00 45 00 00 54 2b 4b 40 00 40 01
00:02:31:927375: l2_lisp_gpe100-output
  l2_lisp_gpe100 l2_hdr_offset_valid l3_hdr_offset_valid
  flags: N L E V I
  ver_res 22 res 60 next_protocol 80 iid 9388282(8f40fa)
00:02:31:927377: l2_lisp_gpe100-tx
  L2-LISP-GPE-TX: load-balance 8
00:02:31:927379: l2-load-balance
  L2-load-balance: index 8
00:02:31:927379: lisp-cp-lookup-l2
  LISP-CP-LOOKUP: map-resolver: 0.0.0.0 destination eid [100] fa:16:3c:50:8f:40
00:02:31:927384: error-drop
  rx:VirtualEthernet0/0/0
00:02:31:927385: drop
  lisp-cp-lookup-l2: drop

the vpp version is vpp-18.07.1-1.x86_64 and openstack version is Stein.

please let me know if you more infomation.

[Naveen Joy (najoy)]

From the trace it's unclear as to why the lisp control plane is dropping the packet. I will work on reproducing this issue. In the meantime, can you test with vpp-19.08 and networking-vpp master to see if the issue's still present.

[Naveen Joy (najoy)]

To troubleshoot further, run the below VPP commands on both nodes -

1) show one eid-table
You should see both the local and remote IP addresses and corresponding mac-addresses

2) show one l2 arp entries
# You should see the remote VM's ARP entry

[Naveen Joy (najoy)]

Ensure that you've allocated a range of VNIs in your ml2_conf.ini file for GPE.
Do not overlap this with the VLAN ranges available for allocation.

For example:

[ml2_vpp]
gpe_vni_ranges = 65000:66000

[Ian Wells (ijw-ubuntu)]

Abandoning this bug after 5 months without response.