Change default route in docker container

问题

I have a docker container that is connected to two networks, the default bridge and a custom bridge. Via the default, it is linked to another container only in the default network and via the custom bridge, it gets an IP address in local network.

LAN -- [homenet] -- container1 -- [bridge] -- container2

sudo docker network inspect homenet
[{  "Name": "homenet",
    "Scope": "local",
    "Driver": "bridge",
    "EnableIPv6": false,
    "IPAM": {
        "Driver": "default",
        "Options": {},
        "Config": [{ "Subnet": "192.168.130.0/24",
                     "Gateway": "192.168.130.8",
                     "AuxiliaryAddresses": { "DefaultGatewayIPv4": "192.168.130.3" }}]
    },
    "Internal": false,
    "Containers": {
        "$cid1": { "Name": "container",
                   "EndpointID": "$eid1_1",
                   "MacAddress": "$mac1_1",
                   "IPv4Address": "192.168.130.38/24", }
    },
    "Options": { "com.docker.network.bridge.name": "br-homenet" },
    "Labels": {}}]

and bridge:

sudo docker network inspect bridge

[{
    "Name": "bridge",
    "Scope": "local",
    "Driver": "bridge",
    "EnableIPv6": false,
    "IPAM": {
        "Driver": "default",
        "Options": null,
        "Config": [{ "Subnet": "172.17.0.0/16" }]
    },
    "Internal": false,
    "Containers": { 
      "$cid2": {
            "Name": "container2",
            "EndpointID": "$eid2",
            "MacAddress": "$mac2",
            "IPv4Address": "172.17.0.2/16",
            "IPv6Address": "" }, 
      "$cid1": {
            "Name": "container1",
            "EndpointID": "$eid1_2",
            "MacAddress": "$mac1_2",
            "IPv4Address": "172.17.0.3/16",
            "IPv6Address": "" }
    },
    "Options": {
        "com.docker.network.bridge.default_bridge": "true",
        "com.docker.network.bridge.enable_icc": "true",
        "com.docker.network.bridge.enable_ip_masquerade": "true",
        "com.docker.network.bridge.host_binding_ipv4": "0.0.0.0",
        "com.docker.network.bridge.name": "docker0",
        "com.docker.network.driver.mtu": "1500"
    },
    "Labels": {}
}]

This works pretty well from the internal network, however, I have a routing problem:

sudo  docker exec -it container1 route -n

Kernel IP routing table
Destination     Gateway         Genmask         Flags Metric Ref    Use Iface
0.0.0.0         172.17.0.1      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 eth0
192.168.130.0   0.0.0.0         255.255.255.0   U     0      0        0 eth1

How can I change the default route to 192.169.130.3 such that it persists a restart?

I can change it while container1 is running with

 pid=$(sudo docker inspect -f '{{.State.Pid}}' container1)
 sudo mkdir -p /var/run/netns
 sudo ln -s /proc/$pid/ns/net /var/run/netns/$pid
 sudo ip netns exec $pid ip route del default 
 sudo ip netns exec $pid ip route add default via 192.168.130.3

but that is gone after a restart. How can I change that?

Update: Apparently, the lexicographical order of the networks could also be part of the issue. I will test it when I get a chance.

回答1:

If I understand the question, the problem is : when restarting a container connected to multiple bridges, how to prefer a bridge to use for default route ?

I searched available options and made some tests, I did not found any docker command line option to specify a default route or to prefer a bridge as default when the container is connected to multiple bridges. When I restart a container connected to the default bridge (bridge) and a custom bridge (your homenet), the default route is automatically set to use the default bridge (gateway 172.17.0.1). This corresponds to the behavior you describe.

Solution 1: Specify a start script in the run command that is in charge to change the default route and start the service(s) you container has to run:

docker run \
  --cap-add NET_ADMIN \ # to allow changing net settings inside the container 
  --name container1 \
  --restart always \ # restart policy
  your_image \
  /path/to/your_start_script.sh

The your_start_script.sh:

ip route del default 
ip route add default via 192.168.130.3

# here goes instructions/services your container is supposed to run

This script has to be available inside the container, it can be on a shared folder (-v option) or loaded at image building with a Dockerfile.

Note: before connecting the container to your custom bridge (docker network connect homenet container1), your_start_script.sh will crash because the default route does not correspond to any available network.

I tested to log the output of ip route inside container1 run with --restart always, after connecting it to the custom bridge it has the wanted default route.

Solution 2: Set container default route from host on container start events

docker events --filter "container=container1" |\
  awk '/container start/ { system("/path/to/route_setting.sh") }'

Where route_setting.sh contains your instructions for changing the container's default route:

pid=$(sudo docker inspect -f '{{.State.Pid}}' container1)
sudo mkdir -p /var/run/netns
sudo ln -s /proc/$pid/ns/net /var/run/netns/$pid
sudo ip netns exec $pid ip route del default 
sudo ip netns exec $pid ip route add default via 192.168.130.3

This solution avoids giving special permissions to the container and transfers the route changing responsibility to the host.

回答2:

nsenter -n -t $(docker inspect --format {{.State.Pid}} $dockername) ip route add something.

nsenter -n -t $(docker inspect --format {{.State.Pid}} $dockername) ip route del something.

回答3:

@Silicium14

Thanks a lot for your 2nd solution. Took me quite long to find a way to set routes upon container start. I changed your lines a bit according to my needs as I need to provide a container name from docker events to the script

First I start the listener for my events.

docker events --filter 'container=box1' --filter 'container=box2' --filter 'event=start' --filter 'event=stop' --format '{{.Actor.Attributes.name}}'|awk '{ system("/work/route_setting.sh " $1) }'

I use more filters as I need the events for two containers of type start or stop Using --format one can control the output very nicely. So only the container name is piped to awk. Which then fires my routing script with the correct containername.

#!/bin/bash

# exit if no container name provided as $1
[ "x$1" = 'x' ] && exit 1
# holds pid of the docker container
pid=''
# read the pid for container
pid=$(docker inspect -f '{{.State.Pid}}' "${1}" 2>/dev/null)
# if for whatevery reason we get pid 0 avoid setting routes
[ "x$pid" = 'x0' ] && pid=''
if [ "x$pid" != 'x' ] ; then
  # let the routing happen 
  mkdir -p /var/run/netns
  ln -s /proc/$pid/ns/net /var/run/netns/$pid
  ip netns exec $pid ip route add 10.0.0.0/8 via 10.66.101.1
  ip netns exec $pid ip route add 192.168.0.0/16 via 10.66.101.1
fi
# clean up broken symlinks which occur when a container is stopped
# verify that your find supports -xtype l
find /var/run/netns -xtype l -exec rm -f '{}' \;