access host's ssh tunnel from docker container

2020-05-18 04:44发布

问题:

Using ubuntu tusty, there is a service running on a remote machine, that I can access via port forwarding through an ssh tunnel from localhost:9999.

I have a docker container running. I need to access that remote service via the host's tunnel, from within the container.

I tried tunneling from the container to the host with -L 9000:host-ip:9999 , then accessing the service through 127.0.0.1:9000 from within the container fails to connect. To check wether the port mapping was on, I tried nc -luv -p 9999 # at host nc -luv -p 9000 # at container

following this, parag. 2 but there was no perceived communication, even when doing nc -luv host-ip -p 9000 at the container

I also tried mapping the ports via docker run -p 9999:9000 , but this reports that the bind failed because the host port is already in use (from the host tunnel to the remote machine, presumably).

So my questions are

1 - How will I achieve the connection? Do I need to setup an ssh tunnel to the host, or can this be achieved with the docker port mapping alone?

2 - What's a quick way to test that the connection is up? Via bash, preferably.

Thanks.

回答1:

I think you can do it by adding --net=host to your docker run. But see also this question: Forward host port to docker container



回答2:

Using your hosts network as network for your containers via --net=host or in docker-compose via network_mode: host is one option but this has the unwanted side effect that (a) you now expose the container ports in your host system and (b) that you cannot connect to those containers anymore that are not mapped to your host network.

In your case, a quick and cleaner solution would be to make your ssh tunnel "available" to your docker containers (e.g. by binding ssh to the docker0 bridge) instead of exposing your docker containers in your host environment (as suggested in the accepted answer).

Setting up the tunnel:

For this to work, retrieve the ip your docker0 bridge is using via:

ifconfig

you will see something like this:

docker0   Link encap:Ethernet  HWaddr 03:41:4a:26:b7:31  
          inet addr:172.17.0.1  Bcast:172.17.255.255  Mask:255.255.0.0

Now you need to tell ssh to bind to this ip to listen for traffic directed towards port 9000 via

ssh -L 172.17.0.1:9000:host-ip:9999

Without setting the bind_address, :9000 would only be available to your host's loopback interface and not per se to your docker containers.

Side note: You could also bind your tunnel to 0.0.0.0, which will make ssh listen to all interfaces.

Setting up your application:

In your containerized application use the same docker0 ip to connect to the server: 172.17.0.1:9000. Now traffic being routed through your docker0 bridge will also reach your ssh tunnel :)

For example, if you have a "DOT.NET Core" application that needs to connect to a remote db located at :9000, your "ConnectionString" would contain "server=172.17.0.1,9000;.



回答3:

on MacOS (tested in v19.03.2),

1) create a tunnel on host

ssh -i key.pem username@jump_server -L 3336:mysql_host:3306 -N

2) from container, you can use host.docker.internal or docker.for.mac.localhost or docker.for.mac.host.internal to reference host.

example,

mysql -h host.docker.internal -P 3336 -u admin -p

note from docker-for-mac official doc

I WANT TO CONNECT FROM A CONTAINER TO A SERVICE ON THE HOST

The host has a changing IP address (or none if you have no network access). From 18.03 onwards our recommendation is to connect to the special DNS name host.docker.internal, which resolves to the internal IP address used by the host. This is for development purpose and will not work in a production environment outside of Docker Desktop for Mac.

The gateway is also reachable as gateway.docker.internal.



回答4:

I'd like to share my solution to this. My case was as follows: I had a PostgreSQL SSH tunnel on my host and I needed one of my containers from the stack to connect to a database through it.

I spent hours trying to find a solution (Ubuntu + Docker 19.03) and I failed. Instead of doing voodoo magic with iptables, doing modifications to the settings of the Docker engine itself I came up with a solution and was shocked I didn't thought of this earlier. The most important things was I didn't want to use the host mode, security first.

Instead of trying to allow a container to talk to the host, I simply added another service to the stack, which would create the tunnel, so other containers could talk to easily without any hacks.

After configuring a host inside my ~/.ssh/config:

Host project-postgres-tunnel
    HostName remote.server.host
    User sshuser
    Port 2200
    ForwardAgent yes
    TCPKeepAlive yes
    ConnectTimeout 5
    ServerAliveCountMax 10
    ServerAliveInterval 15

And adding a service to the stack:

  postgres:
    image: cagataygurturk/docker-ssh-tunnel:0.0.1
    volumes:
      - $HOME/.ssh:/root/ssh:ro
    environment:
      TUNNEL_HOST: project-postgres-tunnel
      REMOTE_HOST: localhost
      LOCAL_PORT: 5432
      REMOTE_PORT: 5432
    # uncomment if you wish to access the tunnel on the host
    #ports:
    #  - 5432:5432

The PHP container started talking through the tunnel without any problems:

postgresql://user:password@postgres/db?serverVersion=11&charset=utf8

Just remember to put your public key inside that host if you haven't already:

ssh-copy-id project-postgres-tunnel

I'm pretty sure this will work regardless of the OS used (MacOS / Linux).



回答5:

On my side, running Docker in Windows Subsystem for Linux (WSL v1), I couldn't use docker0 connection approach. host.docker.internal also doesn't resolve (latest docker version).

However, I found out I could directly use the host-ip insider my docker container.

  1. Get your Host IP (Windows cmd: ipconfig), e.g. 192.168.0.5
  2. Bash into your Container and test if you can ping your host ip:
    - docker exec -it d6b4be5b20f7 /bin/bash
    - apt-get update && apt-get install iputils-ping
    - ping 192.168.0.5
PING 192.168.0.5  (192.168.0.5) 56(84) bytes of data.
64 bytes from 192.168.0.5 : icmp_seq=1 ttl=37 time=2.17 ms
64 bytes from 192.168.0.5 : icmp_seq=2 ttl=37 time=1.44 ms
64 bytes from 192.168.0.5 : icmp_seq=3 ttl=37 time=1.68 ms

Apparently, in Windows, you can directly connect from within containers to the host using the official host ip.



回答6:

My 2 cents for Ubuntu 18.04 - a very simple answer, no need for extra tunnels, extra containers, extra docker options or exposing host.

Simply, when creating a reverse tunnel make sure ssh binds to all interfaces as, by default, it binds ports of the reverse tunnel to localhost only. For example, in putty make sure that option Connection->SSH->Tunnels Remote ports do the same (SSH-2 only) is ticked. This is more or less equivalent to specifying the binding address 0.0.0.0 for the remote part of the tunnel (more details here):

-R [bind_address:]port:host:hostport

However, this did not work for me unless I allowed the GatewayPorts option in my sshd server configuration. Many thanks to Stefan Seidel for his great answer.

In short: (1) you bind the reverse tunnel to 0.0.0.0, (2) you let the sshd server to accept such tunnels.

Once this is done I can access my remote server from my docker containers via the docker gateway 172.17.0.1 and port bound to the host.