HGKeeper is an server for mercurial repositories. It provides access control for SSH access and public HTTP access via hgweb.
It was originally designed to be run in a container but recently support has been added to run it from an existing openssh-server.
HGKeeper is licensed under the GNU AFFERO GENERAL PUBLIC LICENSE version 3.
This project is brand new and only lightly tested. If you find issues, or are looking for ways to help, please check out our issue tracker.
This project needs to generate some source files, so you'll need to first install the embedding tool esc.
$ go get -u github.com/mjibson/esc
Once esc is installed, make sure it's available on your path.
Then you can build hgkeeper with the following commands:
$ go generate ./... $ go build
The initial setup of HGKeeper has a few steps. Since HGKeeper is an SSH server
you will need to generate host keys for it, as well as create the initial
hgkeeper repository which contains the configuration for your install.
You can generate SSH host keys for whatever type you like, but rsa will cover just about everyone. That said, a lot of people prefer to use ed25519 as well.
By default the SSH host keys will be looked for in the directory
the current working directory. This can be changed with the
-H command line arguments to hgkeeper.
This directory will be read and files in it will attempt to be loaded into the server.
To generate a host key you can use the following command, note that you can
create other types via the
-t option, but you should read the
documentation as other options are avaiable for each type.
$ ssh-keygen -t rsa -b 4096 -o host-keys/ssh_host_rsa_key
Before you can run the server we need to create the
repository. This can be done via
hgkeeper setup. You will need to pass the
--admin-username which is the name of the admin user, as well as
--admin-pubkey which is the path to the SSH public key for the new admin
user. By default this will create a new repository under
There are some additional options which you can discover via
hgkeeper setup --help.
hgkeeper has a couple of modes of operation but
serve is the main mode.
setup command is used to bootstrap HGKeeper. It will create the
directory for the repositories as well as the
After initial setup, please make sure to read the
README.md in the
admin repository that was created, as it details how access control works.
serve command is the main mode of operation which is to provide access to
Access control is defined in the
hgkeeper admin repository that is created
hgkeeper setup command. It is implemented via
casbin using the RBAC with deny-override model as a
base. More information can be found in the files
that are placed in the
hgkeeper admin repository.
HGKeeper is available on docker hub under rwgrim/hgkeeper and is updated via CI.
Just like running locally, running in the container is going to require at
least one ssh host key and an
hgkeeper admin repository.
For the rest of these instructions we are going to assume that you have your
ssh host keys in a directory named
host-keys in the current working
Once you have your ssh host keys generated you can create the
repository by running the container with an overridden command.
An extra step to this is that you'll need to volume mount in a file containing
the public key of the initial administrator of this instance. In the
following example we assume that that key is in
Also, since this container is just used for initialization of files on the
host, we're passing the
--rm flag to make sure it's deleted when done.
We can now run the initialization setup that follows.
docker run --rm \ -v $(pwd)/repos:/repos \ -v ~/.ssh/id_rsa.pub:/admin-pubkey:ro \ -e HGK_ADMIN_USERNAME=my_username \ -e HGK_ADMIN_PUBKEY=/admin-pubkey \ -e HGK_REPOS_PATH=/repos \ docker.io/rwgrim/hgkeeper:latest \ hgkeeper setup
Once this step is done you should now have a
repos directory in your current
working directory and it should have a brand new
hgkeeper admin repository
Now that your
hgkeeper admin repository is all ready to go you can run
HGKeeper in its normal
The following example uses the same assumptions as the setup container above,
but it is also going to expose the container on the host's network interface.
The following example runs the container in the background, gives it a name of
hgkeeper, and sets it to always restart. If you're just doing testing, you
will probably want to remove the
--name hgkeeper (name),
docker run -d --name hgkeeper \ --restart=always \ -v $(pwd)/host-keys:/host-keys:ro \ -v $(pwd)/repos:/repos \ -e HGK_SSH_HOST_KEYS=/host-keys \ -e HGK_REPOS_PATH=/repos \ -p 8080:8080 \ -p 22222:22222 \ docker.io/rwgrim/hgkeeper:latest \ hgkeeper serve
And that's it! You can now access your instance via the hosts IP address or DNS and you're good to go!
Of course, you'll probably want to add some more users. To find out how to do
that, be sure to read the
README.md in the
hgkeeper admin repository.
Once the SSH host keys and the
hgkeeper admin repository are created, you
can run HGKeeper with
hgkeeper serve. There are some other options that are
available so be sure to check out
hgkeeper serve --help.
There are a number of steps to integrate HGKeeper with OpenSSH Server and some of them vary across operating systems. If you run into a case where these instructions do not work for you, please reach out via the issue tracker so we can help you and fix the documentation.
When it comes to integrating with OpenSSH server you need to create a user that
will run HGKeeper during SSH connections and own the repositories on disk. This
user can be named whatever you like, but for the purposes of this documentation
we will be naming the user
For most Linux distributions, you can create the
hg user with the following
useradd --home-dir /var/lib/hg --create-home --system --shell /usr/sbin/nologin
To get HGKeeper fully running, you will need to run
hgkeeper setup to create
hgkeeper admin repository as well as the initial admin user. Once this
repository is created, make sure it and it's parent directory are owned by the
hg user and that the
hg user has write permission.
OpenSSH server has some very specific requirements for calling applications
directly. These requirements are that the executable as well as all of the
directories leading up to the executable must be owned by root and not
writeable by the group or other users. To deal with this, we will be
installing HGKeeper into
/usr/local/bin. This directory should fulfill all
of the those requirements. So just
sudo cp hgkeeper /usr/local/bin/ and
make sure that it is owned by
root with a file mode of
The OpenSSH configuration is actually quite easy, you just need to drop the
following snippet into
/etc/ssh/sshd_config. Of course, if you customized
the install location or user name you'll have to adjust that in the snippet
below. Note that the value for
--repos-path needs to be the absolute path
to your repositories.
You may be able to use
/etc/ssh/sshd_config.d/hgkeeper.conf but in our
testing on Debian unstable we were unable to get it working.
Match User hg AuthorizedKeysCommand /usr/local/bin/hgkeeper --repos-path=<path to your repositories> authorized-keys %f AuthorizedKeysCommandUser hg
Once you've created the file, you'll need to reload OpenSSH Server. This is
usually done via
service ssh reload but may vary based on your operating
Once you've reloaded OpenSSH Server you should be able to clone the
admin repository with
hg clone ssh://hg@yourhostname/hgkeeper as long the
machine you're cloning from has the private key that you added as your initial
Creating a new repository in HGKeeper is just the same as using mercurial over
SSH. That is, you can use the
hg init command with a path to HGKeeper. As
long as you have init permission for the path
HGKeeper will create an empty
repository for you.
For example, if you wanted to create a new repository named
myteam/new-project-1, you would do so with
hg init ssh://hg@my-hgkeeper-domain/myteam/new-project-1. If there were no
issues, the command will exit without error and you are now free to clone it
hg clone ssh://hg@my-hgkeeper-domain/myteam/new-project-1.
Fedora 33 disabled
ssh-rsa by default in openssh-client. This breaks the
x/crypto/ssh library. See
this bug for more details.
The current work around for this is to add a host block in your ssh client
configuration to re-enable
ssh-rsa. See the following example:
Host <hgkeeper hostname> PubkeyAcceptedKeyTypes ssh-rsa