Skip to content

About

Re-Connectable secure remote shell

Topics

Resources

Stars

3.9k stars

Watchers

23 watching

Forks

Latest commit

 

History

1,144 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Eternal Terminal

Eternal Terminal is a remote shell that automatically reconnects without interrupting the session.

Website: https://mistertea.github.io/EternalTerminal/.

Packaging status

Packaging status

Installing

macOS

The easiest way to install etis by using Homebrew:

brew install et

If the install fails on including csignal, see #662 (comment)

Then if you want a daemon to launch etserver on every boot:

On m1 (Apple Silicon) Macs:

sudo sed 's:/usr/local/bin/etserver:/opt/homebrew/bin/etserver:g' ../init/launchd/homebrew.mxcl.et.plist | sudo tee /Library/LaunchDaemons/homebrew.mxcl.et.plist
sudo launchctl load -w /Library/LaunchDaemons/homebrew.mxcl.et.plist

On x86 Macs:

sudo cp ../init/launchd/homebrew.mxcl.et.plist /Library/LaunchDaemons/homebrew.mxcl.et.plist
sudo launchctl load -w /Library/LaunchDaemons/homebrew.mxcl.et.plist

Alternatively, a package is available in MacPorts:

sudo port install et

Ubuntu

For Ubuntu, use our PPA:

sudo add-apt-repository ppa:jgmath2000/et
sudo apt-get update
sudo apt-get install et

Or see "Debian/Ubuntu" below to install and build from source (e.g., for ARM).

Debian

For Debian, use our deb repo:

echo "deb [signed-by=/etc/apt/keyrings/et.gpg] https://mistertea.github.io/debian-et/debian-source/ $(grep VERSION_CODENAME /etc/os-release | cut -d= -f2) main" | sudo tee -a /etc/apt/sources.list.d/et.list
sudo mkdir -m 0755 -p /etc/apt/keyrings # only if you're using Debian 11 or older
curl -sSL https://github.com/MisterTea/debian-et/raw/master/et.gpg | sudo tee /etc/apt/keyrings/et.gpg >/dev/null
sudo apt update
sudo apt install et

CentOS 7

Up to the present day the only way to install is to build from source.

CentOS 8

sudo dnf install epel-release
sudo dnf install et

FreeBSD

On FreeBSD, use:

pkg install eternalterminal

Fedora (version 29 and later):

sudo dnf install et

openSUSE

zypper ar -f obs://network
zypper ref
zypper in EternalTerminal

Other Linux

Install dependencies:

  • Fedora (tested on 25):

    sudo dnf install boost-devel libsodium-devel protobuf-devel \
    	protobuf-compiler cmake gflags-devel libcurl-devel
  • Gentoo:

    sudo emerge dev-libs/boost dev-libs/libsodium \
    	dev-libs/protobuf dev-util/cmake dev-cpp/gflags

Download and install from source:

git clone --recurse-submodules --depth 1 https://github.com/MisterTea/EternalTerminal.git
cd EternalTerminal
mkdir build
cd build
cmake ../
make
sudo make install

NixOS

If you use flakes, you can import the bundled NixOS module and let it manage the package, /etc/et.cfg, and the etserver service:

{
  inputs.et.url = "github:MisterTea/EternalTerminal";

  outputs = { nixpkgs, et, ... }: {
    nixosConfigurations.my-host = nixpkgs.lib.nixosSystem {
      system = "x86_64-linux";
      modules = [
        et.nixosModules.default
        ({ ... }: {
          services.eternalTerminal = {
            enable = true;
            openFirewall = true;
            port = 2022;
            settings.Networking.bind_ip = "0.0.0.0";
          };
        })
      ];
    };
  };
}

Windows

Eternal Terminal works under WSL (Windows Subsystem for Linux). Follow the ubuntu instructions.

Docker Image

See docker/README.md

Verifying

Verify that the client is installed correctly by looking for the et executable: which et.

Verify that the server is installed correctly by checking the service status: systemctl status et. On some operating systems, you may need to enable and start the service manually: sudo systemctl enable --now et.

You are ready to start using ET!

Installed programs

An Eternal Terminal installation may provide these executables:

  • et is the command-line client. It uses SSH to authenticate and start the remote session, then maintains the reconnectable ET connection.
  • etserver is the system service that accepts ET client connections and routes them to the correct user's session. Its default TCP port is 2022.
  • etterminal is an internal server-side helper launched through SSH. Normal users do not invoke it directly.
  • htm is the foreground client for HTM, the bundled terminal multiplexer. It speaks tmux control mode so compatible terminal emulators can show native windows, tabs, and splits.
  • htmd is the per-user HTM daemon. It owns persistent multiplexer sessions and is started automatically by htm when needed.

For the protocol-level relationship between et, etserver, and etterminal, see the protocol documentation. For HTM, see the HTM design documentation.

Configuring

If you'd like to modify the server settings (e.g. to change the listening port), edit /etc/et.cfg. On NixOS, set services.eternalTerminal.settings instead so the generated /etc/et.cfg stays in sync with your system configuration.

Using

ET uses ssh for handshaking and encryption, so you must be able to ssh into the machine from the client. Make sure that you can ssh user@hostname.

ET uses TCP, so you need an open port on your server. By default, it uses 2022.

Once you have an open port, the syntax is similar to ssh. Username is default to the current username starting the et process, use -u or user@ to specify a different one if necessary.

et hostname (etserver running on default port 2022, username is the same as current)
et user@hostname:8000 (etserver running on port 8000, different user)

Saved sessions

On macOS and Linux, et saves each direct session's reattachment credentials in ~/.et/sessions (owner-only, plaintext), so after a client crash or reboot you can return to a remote shell that is still running. Sessions get a generated YYYYMMDD-xxxx name unless you pick one; --attach and --kill accept a name or a unique substring of the name or terminal title. Port forwards, agent forwarding, and jumphosts are not restored on attach.

et --name work hostname   # start (or reattach to) a named session
et --list                 # list saved sessions without contacting servers
et --attach work          # reattach after the client restarted
et --kill work            # end the remote session and remove its record
et --no-persist hostname  # do not write credentials to disk

You can specify a jumphost and the port et is running on jumphost using --jumphost and --jport. If no --jport is given, et will try to connect to default port 2022.

et hostname -jumphost jump_hostname (etserver running on port 2022 on both hostname and jumphost)
et hostname:8888 --jumphost jump_hostname --jport 9999

Additional arguments that et accepts are port forwarding pairs with --tunnel "18000:8000, 18001-18003:8001-8003" (or OpenSSH-style -L), and a command to run immediately after the connection is set up through --command or as a positional command after the host. Short flags match OpenSSH: -p is the sshd port, --port is the etserver port, and -t requests a pty.

Starting from the latest release, et supports parsing both user-specific and system-wide SSH config files. The config file is required when your sshd on server/jumphost is listening on a port which is not 22. Here is an example SSH config file showing how to setup when

  • there is a jumphost in the middle
  • sshd is listening on a port that is not 22
  • connecting to a different username other than the current one.
Host dev
  HostName 192.168.1.1
  User fred
  Port 5555
  ProxyJump user@jumphost.example.org:22

With the ssh config file set as above, you can simply call et with

et dev (etserver running on port 2022 on both hostname and jumphost)
et dev:8000 -jport 9000 (etserver running on port 9000 on jumphost)

To isolate ET from ambient SSH configuration, pass an absolute path with --ssh-config. ET reads only that file for its own destination and jumphost lookup and passes the same file to every SSH process it starts, including the implicit ProxyJump connection. The file must be readable, regular, and not a symbolic link. The path must be absolute (/path on Unix; C:\path, C:/path, or a UNC path on Windows) and may contain only ASCII letters, digits, /, ., _, and - (plus : and \ on Windows). Spaces are not allowed: OpenSSH does not quote this path when constructing its implicit ProxyJump command. Pass --ssh-config none or the equivalent --no-ssh-config to disable both user and system SSH configuration entirely. The two options are mutually exclusive.

VS Code and Cursor (Remote-SSH)

ET can stand in for ssh as the transport for VS Code's and Cursor's Remote-SSH extensions, so a remote window rides out laptop sleep and network changes instead of prompting you to reload. Both ends need a recent ET: the client must include #874, and the server's etserver and etterminal must support et -T (the 7.0.0 packages do not). ssh hostname must already work without a password prompt (use a key).

Point the editor at et1, which is installed next to et. It runs et --close-on-hangup --disconnect-timeout=10080 "$@": the session is kept on the server through a week-long disconnect (the value is in minutes) and ended when the editor closes the connection. Remote-SSH only accepts a path to a single executable, so if you need more et options (for example --port for an etserver not on the default port), copy et1 and add them there.

For VS Code, add to settings.json (use the output of command -v et1 as the path):

{
  "remote.SSH.path": "/usr/local/bin/et1",
  "remote.SSH.useLocalServer": false,
  "remote.SSH.reconnectionGraceTime": 604800
}
  • remote.SSH.useLocalServer: false is required. In the default local-server mode, VS Code kills the transport process a few seconds after it stops responding, which throws away the ET client that would have reconnected.
  • remote.SSH.reconnectionGraceTime (seconds) is how long the VS Code server on the remote keeps your window's state while disconnected. The default is three hours. It only takes effect when that server starts, so run "Remote-SSH: Kill VS Code Server on Host..." after changing it.

For Cursor, set remote.SSH.path to et1 as well. Cursor's Remote-SSH extension has no local-server mode or grace-time setting; its closest equivalent is remote.SSH.serverShutdownTimeout (seconds, default 300), which controls how long the Cursor server stays up after the last client disconnects.

Programmatic control (etctl)

Normally et drives a terminal for a human. et --ctl instead backgrounds a session with no local terminal and serves it on a per-user unix socket, and the companion etctl command drives that session by name from a script or program. You get a command's clean output and real exit code, instead of scraping a rendered screen.

et --ctl --name main user@hostname   # background a named session (idempotent via: etctl open main user@hostname)
etctl run    main 'uname -a'         # run a command; captured output + the real exit code
etctl read   main                    # retained output (non-destructive; --cursor N to resume)
etctl expect main 'Password:'        # wait for a prompt
etctl writeln main --secret          # answer it without echoing the password
etctl key     main eof               # send Ctrl-D to end the session cleanly

etctl open is an idempotent wrapper over et --ctl. HOST comes first, and any further et arguments follow it, so etctl open main user@hostname --command '<cmd>' runs a startup command on connect (for example, to drop into a bare, prompt-free shell for cleaner capture). The session's remote shell is stamped with ETCTL_SESSION=<name>, so a process can tell which named control session it is running under. The socket lives at ~/.et/control/<name>.sock (0700 dir, 0600 socket, owning-uid only), beside the session's saved record in ~/.et/sessions/<name>.

How run frames a command

run has to know where a command's output starts and ends and what it exited with, and it has to inject the command safely (a multi-line body must run as one command, and quotes/braces/!/parse errors must not desync the frame or hang on a continuation prompt). It picks the cleanest of three framings for the far-side prompt, detected once per session and cached (~/.et/control/<name>.framing):

  • Bracketed paste + OSC 133 (the default when the prompt has FinalTerm/iTerm2 shell integration and a bracketed-paste-aware line editor -- the official bash, zsh, fish, and xonsh integrations all qualify). The bare command is sent inside bracketed paste, so the real command shows in the scrollback with no wrapper and no injected markers, and the output boundaries + exit code come straight from the prompt's own OSC 133 C/D marks.
  • Eval here-doc + OSC 133 (a prompt with OSC 133 but no bracketed paste, e.g. tcsh). The body is wrapped in eval "$(cat <<'BODY' ... )" -- so it is handed to the shell as data, immune to its own syntax -- and boundaries come from OSC 133. No echo markers, but the wrapper is visible.
  • Eval here-doc + echo markers (the universal fallback for a shell with no OSC 133, e.g. dash). The eval is bracketed by echo <mark> ... <mark>:$?.

run also trims the prompt-prep sequences a shell splices around the output (a zsh/fish end-of-line mark, iTerm2's OSC 1337 context report, a title, bracketed-paste toggles) so the captured output is the command's alone, and it disables history expansion on the control session (a script-driven session has no use for interactive !). Force a framing with --framing=osc133 (bracketed paste) or --framing=mark (echo markers); the default is --framing=auto.

Run etctl with no arguments for the full verb list, etctl <verb> --help for any verb's options, and etctl --version for the version. --ctl is not available on Windows.

Building from Source

macOS

To build Eternal Terminal on Mac, install its dependencies with Homebrew:

brew install autoconf automake libtool
git clone --recurse-submodules --depth 1 https://github.com/MisterTea/EternalTerminal.git
cd EternalTerminal
mkdir build
cd build
cmake ../
make -j$(nproc) && sudo make install

To run an et server for testing, run ./etserver. To run an et server daemon persistently across reboots:

sudo cp ../init/launchd/homebrew.mxcl.et.plist /Library/LaunchDaemons
sudo launchctl load -w /Library/LaunchDaemons/homebrew.mxcl.et.plist

Debian/Ubuntu

Grab the deps and then follow this process.

Debian/Ubuntu Dependencies:

sudo apt install libsodium-dev autoconf libtool \
	libprotobuf-dev protobuf-compiler libutempter-dev libcurl4-openssl-dev \
    build-essential ninja-build cmake git zip pkg-config

Fetch source, build and install:

git clone --recurse-submodules --depth 1 https://github.com/MisterTea/EternalTerminal.git
cd EternalTerminal
mkdir build
cd build
# For ARM (including OS/X with apple silicon):
if [[ $(uname -a | grep 'arm\|aarch64') ]]; then export VCPKG_FORCE_SYSTEM_BINARIES=1; fi
cmake -DCPACK_GENERATOR=DEB ../
make -j$(nproc) package
sudo dpkg --install *.deb

Once built, the binary only requires libprotobuf-dev.

Disable et server by sudo systemctl disable --now et

CentOS 7

Install dependencies:

sudo yum install epel-release
sudo yum install cmake3 boost-devel libsodium-devel protobuf-devel \
     protobuf-compiler gflags-devel protobuf-lite-devel libcurl-devel \
     perl-IPC-Cmd perl-Data-Dumper libunwind-devel libutempter-devel

Install scl dependencies

sudo yum install centos-release-scl
sudo yum install devtoolset-11 devtoolset-11-libatomic-devel rh-git227

Download and install from source (see #238 for details):

git clone --recurse-submodules --depth 1 https://github.com/MisterTea/EternalTerminal.git
cd EternalTerminal
mkdir build
cd build
scl enable devtoolset-11 rh-git227 'cmake3 ../'
scl enable devtoolset-11 'make && sudo make install'
sudo cp ../systemctl/et.service /etc/systemd/system/
sudo cp ../etc/et.cfg /etc/

Find the actual location of et:

which etserver

Correct the service file (see #180 for details).

sudo sed -ie "s|ExecStart=[^[:space:]]*[[:space:]]|ExecStart=$(which etserver) |" /etc/systemd/system/et.service

Alternatively, open the file /etc/systemd/system/et.service in an editor and correct the ExectStart=... line to point to the correct path of the etserver binary.

ExecStart=/usr/local/bin/etserver --cfgfile=/etc/et.cfg --logtostdout

Reload systemd configs:

sudo systemctl daemon-reload

Start the et service:

sudo systemctl enable --now et.service

Building using Docker

Builder Dockerfiles are located at deployment/. Supported OSes: CentOS 8, openSUSE and Ubuntu.

Reporting issues

If you have any problems with installation or usage, please file an issue on GitHub.

Server/Client Overview

Eternal Terminal uses three binaries:

  • et (client): Runs on the user's machine (client). Connects to a remote server over SSH to launch the terminal session, then connects to etserver on port 2022 for the persistent session.
  • etterminal (server-side user process): Runs on the server as the user (launched by et via SSH). Hosts the terminal session and connects to etserver via a FIFO to register the session.
  • etserver (server daemon): Runs permanently on the server (usually as root/system service). Listens on TCP port 2022 (default) and manages connections between et clients and etterminal sessions.

Which machines need each part:

  • Client machine: needs et.
  • Server machine: needs etserver (service/demon) and etterminal (installed for users; launched by et over SSH).

Port: By default etserver listens on TCP 2022 (configured in /etc/et.cfg via the Networking.port setting).

Service setup: After installation, start/enable etserver via systemd (systemctl enable --now et) or launchd (macOS), or run ./etserver for testing.

Protocol and design documentation

Developers

About

Re-Connectable secure remote shell

Topics

Resources

Stars

3.9k stars

Watchers

23 watching

Forks

Releases

Sponsor this project

Packages

Used by

Contributors

Languages