Is this an Asymmetric Routing routing issue?

johnpoz · Apr 20, 2021, 12:42 PM

Well if your seeing the traffic go out, and its tagged correctly and to the right mac.. It has zero to do with pfsense..

You don't have any static mac setup in pfsense do you.. Maybe you setup static arp for that mac, and its changed?

Helmut101 · Apr 20, 2021, 12:46 PM

@johnpoz said in Is this an Asymmetric Routing routing issue?:

Maybe you setup static arp for that mac, and its changed?

I do have static ARP/MAC - the container gets its IP using DHCP and this is assigned based on MAC. However, checked and MAC is Ok (and I can also reach the container from pfsense/OpenVPN, or from the Management LAN 10, just not from the vlan 30).

I am out now today. Many thanks so far, this is really helpful and, while I am not yet further with my problem, I am learning a lot!

johnpoz · Apr 20, 2021, 12:54 PM

@helmut101 said in Is this an Asymmetric Routing routing issue?:

just not from the vlan 30).

That sure doesn't make any sense.. You sure you don't have a firewall this thing your trying to reach, or odd routing for 30 network on your dest device? So nothing in 30 can talk to it, but 30 can talk to other devices in the 40 vlan..

Can you sniff on 40.17 and validate it actually sees the traffic?

Helmut101 · Apr 21, 2021, 3:03 AM

@johnpoz said in Is this an Asymmetric Routing routing issue?:

ewall this thing your trying t

Yes, tomorrow I have a bit more time. I will look into this carefully and test more siffing at different points, including the VM itself. Will report back.

If it wasn't that strange I would have not written here.. I was working on this issue for 4 days so far.

Helmut101 · Apr 21, 2021, 3:08 AM

Alright, so this will get long.

TL;DR

I currently do not know why, but on the specific Host, there was/is a bridge ethernet link / virtual nic configured that forwarded outgoing routes to the wrong subnet (192.168.16.0) - I have never heard of this subnet and I don't know why this ip/bridge/link ended up there.

I solved the issue (for the moment) with:

ifconfig br-985a84259068 down

But: Once the VM is restarted, the bridge appears again. I am still working on this.

Sleuthing (long)

This was a long walk down the rabbit hole. But I'll write here, perhaps someone else will find any of the commands useful for similar catch the rabbit tasks.

Here's setup for testing:

60 is my IOT subnet
40 is my Sevrice subnet
30 is my Consumer subnet
Client 30.11, where 30 is the subnet/vlan and 11 the IP
Host 40.17, issue getting reached from subnet 30 clients
Host 40.8 no issue getting reached, can reach 40.17
Host 60.10 no issue getting reached, can reach 40.17

This already is really strange. In addition, I could reach 40.17 just
fine from pfsense (ping) and when connected through OpenVPN.

1. Check Routing

On VM 40.17

ip route

default via 192.168.40.1 dev eth0
172.17.0.0/16 dev docker0 proto kernel scope link src 172.17.0.1
172.23.0.0/16 dev br-5acdb2ca8271 proto kernel scope link src 172.23.0.1 linkdown
172.28.0.0/16 dev br-2d547cdc7389 proto kernel scope link src 172.28.0.1
192.168.16.0/20 dev br-985a84259068 proto kernel scope link src 192.168.16.1
192.168.40.0/24 dev eth0 proto kernel scope link src 192.168.40.17

The default route looks fine, but why are there other routes?

Compare output to other Host 40.8 without issues:

default via 192.168.40.1 dev eth0
192.168.40.0/24 dev eth0 proto kernel scope link src 192.168.40.8

The 172 routes may be explained with Docker running on 40.17,
but 192.168.16.0/20 looks strange.

2. Checks Packets (tcpdump)

Now, as suggested, check whether packets really arrive at the host.

40.17:

tcpdump 'host 192.168.30.11 and port not 22'

30.11:

ping 192.168.40.17

Output (tcpdump):

01:50:58.811226 IP 192.168.30.11 > 192.168.40.17: ICMP echo request, id 1221, seq 1, length 64
01:50:59.816840 IP 192.168.30.11 > 192.168.40.17: ICMP echo request, id 1221, seq 2, length 64
01:51:00.820305 IP 192.168.30.11 > 192.168.40.17: ICMP echo request, id 1221, seq 3, length 64
01:51:01.823602 IP 192.168.30.11 > 192.168.40.17: ICMP echo request, id 1221, seq 4, length 64
01:51:02.827368 IP 192.168.30.11 > 192.168.40.17: ICMP echo request, id 1221, seq 5, length 64
01:51:03.831271 IP 192.168.30.11 > 192.168.40.17: ICMP echo request, id 1221, seq 6, length 64

They arrive, but: nothing is returned.

Verify/compare to output of the same commands on working host 40.8:

01:49:39.460155 IP 192.168.30.11 > 192.168.40.8: ICMP echo request, id 1217, seq 1, length 64
01:49:39.460184 IP 192.168.40.8 > 192.168.30.11: ICMP echo reply, id 1217, seq 1, length 64
01:49:40.461106 IP 192.168.30.11 > 192.168.40.8: ICMP echo request, id 1217, seq 2, length 64
01:49:40.461133 IP 192.168.40.8 > 192.168.30.11: ICMP echo reply, id 1217, seq 2, length 64
01:49:41.461886 IP 192.168.30.11 > 192.168.40.8: ICMP echo request, id 1217, seq 3, length 64
01:49:41.461918 IP 192.168.40.8 > 192.168.30.11: ICMP echo reply, id 1217, seq 3, length 64

3. Check routing

At this moment, I was pretty sure to have the issue isolated to the Host 40.17 itself.
Something is going on with the routing.

on host 40.17:

ip route get 192.168.30.11

192.168.30.11 dev br-985a84259068 src 192.168.16.1 uid 0
cache

uh?

compare on working host 40.8:

ip route get 192.168.30.11

192.168.30.11 via 192.168.40.1 dev eth0 src 192.168.40.8 uid 0
cache

Why is outgoing traffic routed through a bridge called br-985a84259068 to subnet 192.168.16.1?

on 40.17:
Check:

cat /etc/network/interfaces

auto lo
iface lo inet loopback

auto eth0
iface eth0 inet dhcp

ok.. further check routes

apt install net-tools
route -n

Kernel IP routing table
Destination Gateway Genmask Flags Metric Ref Use Iface
0.0.0.0 192.168.40.1 0.0.0.0 UG 0 0 0 eth0
172.17.0.0 0.0.0.0 255.255.0.0 U 0 0 0 docker0
172.23.0.0 0.0.0.0 255.255.0.0 U 0 0 0 br-5acdb2ca8271
172.28.0.0 0.0.0.0 255.255.0.0 U 0 0 0 br-2d547cdc7389
192.168.16.0 0.0.0.0 255.255.240.0 U 0 0 0 br-985a84259068 <--- What is this??
192.168.40.0 0.0.0.0 255.255.255.0 U 0 0 0 eth0

Study what are network bridges:
https://wiki.archlinux.org/index.php/Network_bridge
https://tldp.org/HOWTO/BRIDGE-STP-HOWTO/set-up-the-bridge.html

bridge link

8: vetha2e5a47@if7: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 master br-2d547cdc7389 state forwarding priority 32 cost 2
10: vethcd0643c@if9: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 master br-985a84259068 state forwarding priority 32 cost 2 <-- Here it is
14: veth992d5b3@if13: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 master br-2d547cdc7389 state forwarding priority 32 cost 2
16: vethb6721a9@if15: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 master br-2d547cdc7389 state forwarding priority 32 cost 2
18: veth7dfb21f@if17: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 master br-2d547cdc7389 state forwarding priority 32 cost 2
20: vethc3562b4@if19: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 master br-2d547cdc7389 state forwarding priority 32 cost 2
22: veth9017e4e@if21: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 master br-2d547cdc7389 state forwarding priority 32 cost 2

apt install bridge-utils
brctl show br-985a84259068

bridge name bridge id STP enabled interfaces
br-985a84259068 8000.02428b97932d no vethcd0643c <-- Here, too

ifconfig

vethcd0643c: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500
inet6 fe80::dc58:85ff:fef0:eef1 prefixlen 64 scopeid 0x20<link>
ether de:58:85:f0:ee:f1 txqueuelen 0 (Ethernet)
RX packets 8644 bytes 680480 (664.5 KiB)
RX errors 0 dropped 0 overruns 0 frame 0
TX packets 7417 bytes 1041527 (1017.1 KiB)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0

ifconfig vethcd0643c down

Test route again:

ip route get 192.168.30.11
192.168.30.11 dev br-985a84259068 src 192.168.16.1 uid 0
    cache

Also down with the bridge:

ifconfig br-985a84259068 down

192.168.30.11 via 192.168.40.1 dev eth0 src 192.168.40.17 uid 0
cache

Yay!

That is it for the moment. If I restart the LXC container, the bridge is
added again with the same name. Who does this? I do not know yet.

Helmut101 · Apr 21, 2021, 8:06 AM

What a nightmare.. that is finally solved. Ended up in the a complete different hole. It wasn't the rabbit, it was the docker.

TL;DR

On 40.17 LXC, I had docker running with sevral configs.
Docker apparently decides for a default subnet based on a list of ips from:
172.17.0.0/16", 172.18.0.0/16", "172.19.0.0/16",
"172.20.0.0/14", "172.24.0.0/14" "172.28.0.0/14", "192.168.0.0/16"

It verifies whether that IP range is in use. In my case, it somehow missed that 192.168.0.0/16 is near my VLAN config.

Adding

{
    "bip": "193.168.1.5/24",
	"default-address-pools":
	[
		{"base":"172.17.0.0/16","size":24}
	]
}

to /etc/docker/daemon.json solved the problem. But I had to identify first which container used the default network, stop it, reload docker, and start it, to refresh the network.

docker network list

> NETWORK ID     NAME                DRIVER    SCOPE
> 4d17f9cc818b   bridge              bridge    local
> 985a84259068   docker_default      bridge    local<-- this
> 2d547cdc7389   funkwhale_default   bridge    local
> abddd765db3e   host                host      local
> 5acdb2ca8271   iris_default        bridge    local
> be879c14dc73   none                null      local

docker network inspect 985a84259068

>         "Containers": {
>             "24f38ca4c3e1080f050b868f4b980f3616b8047be45809276e74e217bf2f7f57": {
>                 "Name": "Solaranzeige", <--- this
>                 "EndpointID": "8274319e9ec797c19bcd46d2aadf9277d249135a3fbf326abc14b4893f994081",
>                 "MacAddress": "02:42:c0:a8:10:02",
>                 "IPv4Address": "192.168.16.2/20",  <--- this
>                 "IPv6Address": ""
>             }
>         },

docker stop Solaranzeige
systemctl daemon-reload
systemctl restart docker

Verify:

ip addr | grep 192
    inet 192.168.40.17/24 brd 192.168.40.255 scope global eth0

Only the native VLAN!

Helmut101 · Apr 21, 2021, 8:39 AM

Thanks so much to everybody involved here. I was entirely wrong in my initial suspicion, but the analysis helped me better understand how networks work, so I do not consider this as lost time.

Some revelations:

for incoming traffic, wireshark, tcpdump and packet capture (pfsense) are king
for outgoing traffic, ip route get [host ip] helps to see in which direction traffic leaves (or doesn't leave) the host