CODEWITHSUNDEEP
bashgitdocker
crooksey
crooksey

Reputation: 8809

Clone private git repo with dockerfile

I have copied this code from what seems to be various working dockerfiles around, here is mine:

FROM ubuntu

MAINTAINER Luke Crooks "[email protected]"

# Update aptitude with new repo
RUN apt-get update

# Install software 
RUN apt-get install -y git python-virtualenv

# Make ssh dir
RUN mkdir /root/.ssh/

# Copy over private key, and set permissions
ADD id_rsa /root/.ssh/id_rsa
RUN chmod 700 /root/.ssh/id_rsa
RUN chown -R root:root /root/.ssh

# Create known_hosts
RUN touch /root/.ssh/known_hosts

# Remove host checking
RUN echo "Host bitbucket.org\n\tStrictHostKeyChecking no\n" >> /root/.ssh/config

# Clone the conf files into the docker container
RUN git clone [email protected]:Pumalo/docker-conf.git /home/docker-conf

This gives me the error

Step 10 : RUN git clone [email protected]:Pumalo/docker-conf.git /home/docker-conf
 ---> Running in 0d244d812a54
Cloning into '/home/docker-conf'...
Warning: Permanently added 'bitbucket.org,131.103.20.167' (RSA) to the list of known hosts.
Permission denied (publickey).
fatal: Could not read from remote repository.

Please make sure you have the correct access rights
and the repository exists.
2014/04/30 16:07:28 The command [/bin/sh -c git clone [email protected]:Pumalo/docker-conf.git /home/docker-conf] returned a non-zero code: 128

This is my first time using dockerfiles, but from what I have read (and taken from working configs) I cannot see why this doesn't work.

My id_rsa is in the same folder as my dockerfile and is a copy of my local key which can clone this repo no problem.

Edit:

In my dockerfile I can add:

RUN cat /root/.ssh/id_rsa

And it prints out the correct key, so I know its being copied correctly.

I have also tried to do as noah advised and ran:

RUN echo "Host bitbucket.org\n\tIdentityFile /root/.ssh/id_rsa\n\tStrictHostKeyChecking no" >> /etc/ssh/ssh_config

This sadly also doesn't work.

Upvotes: 383

Views: 459526

Answers (12)

Mohit Motwani
Mohit Motwani

Reputation: 4792

What we are doing is using AWS Secret Manager (You can use any that you have access to).

  1. Create an SSH Key and add the public key to an authorised GitHub Account. Perhaps a common GitHub account in your organization that for example a DevOPs Team has access to.
  2. Your private key has new lines. If you copy this directly to your secret manager, the new lines are removed. Therefore, base64 encode your private key with
cat ssh_private_file | base64 -w 0
  1. Add your base64 encoded private key to aws secret manager (or any secret manager).
  2. Let's say the key is called 'ssh_private' inside your secret name "Private_Repo", you can access your private key
aws secretsmanager get-secret-value \
    --secret-id Private_Repo --query "SecretString" --output text --region $region_name \
    | jq -r ".ssh_private" | base64 -d

Make sure you have jq installed.

  1. When you build your image, you need to pass your credentials to the docker build command to authenticate the process to use the secret manager. The following is a snippet of our Dockerfile:
ARG AWS_DEFAULT_REGION
ARG AWS_ACCESS_KEY_ID
ARG AWS_SECRET_ACCESS_KEY
ARG ssh_file=id_rsa

ENV AWS_DEFAULT_REGION=$AWS_DEFAULT_REGION
ENV AWS_ACCESS_KEY_ID=$AWS_ACCESS_KEY_ID
ENV AWS_SECRET_ACCESS_KEY=$AWS_SECRET_ACCESS_KEY

RUN curl "https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip" -o "awscliv2.zip" \
    && unzip awscliv2.zip \
    && ./aws/install \
    && aws secretsmanager get-secret-value \
    --secret-id Private_Repo --query "SecretString" --output text --region $AWS_DEFAULT_REGION \
    | jq -r ".ssh_private" | base64 -d > ${ssh_file} \
    && chmod 0700 id_rsa

RUN eval "$(ssh-agent -s)" \
    && ssh-add ${ssh_file} \
    && mkdir ~/.ssh/ \
    && echo "Host github.com\n\tIdentityFile ${ssh_file}\n\tStrictHostKeyChecking no\n" >> ~/.ssh/config

The above

  • reads the private key from secret manager and writes to a file, -
  • ssh-adds the files
  • edits your ssh config file to access Github.com host with this private key.

You can pass the credentials with the following command if they are saved as environment variables.

DOCKER_BUILDKIT=1 docker build --build-arg AWS_ACCESS_KEY_ID=$env:AWS_ACCESS_KEY_ID --build-arg AWS_SECRET_ACCESS_KEY=$env:AWS_SECRET_ACCESS_KEY --build-arg AWS_DEFAULT_REGION=$env:AWS_DEFAULT_REGION -f docker/Dockerfile .

This approach allows you to delete the private key from your image after all the installations are done and the container doesn't need to access private repos during run time.

I believe, the main downside of this approach is turning off StringHostKeyChecking in your ssh config(I'm not aware of the consequences here) and that secret managers cost some cents.

Upvotes: 1

stacksonstacks
stacksonstacks

Reputation: 9321

recently had a similar issue with a private repository in a Rust project. I suggest avoiding ssh permissions/config altogether.

Instead:

  1. clone the repository within the build environment e.g. CI, where the permissions already exist (or can be easily configured)
  2. copy the files into the Dockerfile (this can also be cached natively within the CI)

Example

part 1) within CI

CARGO_HOME=tmp-home cargo fetch

part 2) within Dockerfile

COPY tmp-home $CARGO_HOME

the process is the same regardless of language/package system

Upvotes: 0

BMitch
BMitch

Reputation: 264861

You often do not want to perform a git clone of a private repo from within the docker build. Doing the clone there involves placing the private ssh credentials inside the image where they can be later extracted by anyone with access to your image.

Instead, the common practice is to clone the git repo from outside of docker in your CI tool of choice, and simply COPY the files into the image. This has a second benefit: docker caching. Docker caching looks at the command being run, environment variables it includes, input files, etc, and if they are identical to a previous build from the same parent step, it reuses that previous cache. With a git clone command, the command itself is identical, so docker will reuse the cache even if the external git repo is changed. However, a COPY command will look at the files in the build context and can see if they are identical or have been updated, and use the cache only when it's appropriate.

BuildKit has a feature just for ssh which allows you to still have your password protected ssh keys, the result looks like:

# syntax=docker/dockerfile:experimental
FROM ubuntu as clone
LABEL maintainer="Luke Crooks <[email protected]>"

# Update aptitude with new repo
RUN apt-get update \
 && apt-get install -y git

# Make ssh dir
# Create known_hosts
# Add bitbuckets key
RUN mkdir /root/.ssh/ \
 && touch /root/.ssh/known_hosts \
 && ssh-keyscan bitbucket.org >> /root/.ssh/known_hosts

# Clone the conf files into the docker container
RUN --mount=type=ssh \
    git clone [email protected]:User/repo.git

And you can build that with:

$ eval $(ssh-agent)
$ ssh-add ~/.ssh/id_rsa
(Input your passphrase here)
$ DOCKER_BUILDKIT=1 docker build -t your_image_name \
  --ssh default=$SSH_AUTH_SOCK .

Again, this is injected into the build without ever being written to an image layer, removing the risk that the credential could accidentally leak out.

BuildKit also has a features that allow you to pass an ssh key in as a mount that never gets written to the image:

# syntax=docker/dockerfile:experimental
FROM ubuntu as clone
LABEL maintainer="Luke Crooks <[email protected]>"

# Update aptitude with new repo
RUN apt-get update \
 && apt-get install -y git

# Make ssh dir
# Create known_hosts
# Add bitbuckets key
RUN mkdir /root/.ssh/ \
 && touch /root/.ssh/known_hosts \
 && ssh-keyscan bitbucket.org >> /root/.ssh/known_hosts

# Clone the conf files into the docker container
RUN --mount=type=secret,id=ssh_id,target=/root/.ssh/id_rsa \
    git clone [email protected]:User/repo.git

And you can build that with:

$ DOCKER_BUILDKIT=1 docker build -t your_image_name \
  --secret id=ssh_id,src=$(pwd)/id_rsa .

Note that this still requires your ssh key to not be password protected, but you can at least run the build in a single stage, removing a COPY command, and avoiding the ssh credential from ever being part of an image.

If you are going to add credentials into your build, consider doing so with a multi-stage build, and only placing those credentials in an early stage that is never tagged and pushed outside of your build host. The result looks like:

FROM ubuntu as clone

# Update aptitude with new repo
RUN apt-get update \
 && apt-get install -y git
# Make ssh dir
# Create known_hosts
# Add bitbuckets key
RUN mkdir /root/.ssh/ \
 && touch /root/.ssh/known_hosts \
 && ssh-keyscan bitbucket.org >> /root/.ssh/known_hosts

# Copy over private key, and set permissions
# Warning! Anyone who gets their hands on this image will be able
# to retrieve this private key file from the corresponding image layer
COPY id_rsa /root/.ssh/id_rsa

# Clone the conf files into the docker container
RUN git clone [email protected]:User/repo.git

FROM ubuntu as release
LABEL maintainer="Luke Crooks <[email protected]>"

COPY --from=clone /repo /repo
...

To force docker to run the git clone even when the lines before have been cached, you can inject a build ARG that changes with each build to break the cache. That looks like:

# inject a datestamp arg which is treated as an environment variable and
# will break the cache for the next RUN command
ARG DATE_STAMP
# Clone the conf files into the docker container
RUN git clone [email protected]:User/repo.git

Then you inject that changing arg in the docker build command:

date_stamp=$(date +%Y%m%d-%H%M%S)
docker build --build-arg DATE_STAMP=$date_stamp .

Upvotes: 105

Mechanic
Mechanic

Reputation: 5380

In addition to @Calvin Froedge's approach to use Personal Access Token (PAT),

You may need to add oauth or oauth2 as username before your PAT to authenticate like this:

https://oauth:<token>@git-url.com/user/repo.git

Upvotes: 0

Neil
Neil

Reputation: 8634

p.s. this solution is quick & easy; but at a cost of reduced security (see comments by @jrh).

Create an access token: https://github.com/settings/tokens

pass it in as an argument to docker (p.s. if you are using CapRover, set it under App Configs)

In your Dockerfile:

ARG GITHUB_TOKEN=${GITHUB_TOKEN}
RUN git config --global url."https://${GITHUB_TOKEN}@github.com/".insteadOf "https://github.com/"
RUN pip install -r requirements.txt

p.s. this assumes that private repos are in the following format in requirements.txt:

git+https://github.com/<YOUR-USERNAME>/<YOUR-REPO>.git

Upvotes: 4

Jerem Lachkar
Jerem Lachkar

Reputation: 1207

Nowsaday you can use the Buildkit option --ssh default when you build your container ; Prior to build, you need to add your SSH deploy key to your ssh-agent.

Here is the full process from the beginning :

  1. Create a key pair on your deployment server. Just run ssh-keygen -t ecdsa Store your key pair into ~/.ssh

  2. Add your public key generated (.pub extension) at your git provider website (gitlab, github..)

  3. Add your key to your ssh-agent (a program that basically manages your keys easier than handling every file)

eval $(ssh-agent)
ssh-add /path/to/your/private/key
  1. Add this to your Dockerfile :
# this 3 first lines add your provider public keys to known_host 
# so git doesn't get an error from SSH.

RUN mkdir -m 700 /root/.ssh && \
  touch -m 600 /root/.ssh/known_hosts && \
  ssh-keyscan your-git-provider.com > /root/.ssh/known_hosts 


# now you can clone with --mount=type=ssh option, 
# forwarding to Docker your host ssh agent

RUN mkdir -p /wherever/you/want/to/clone && cd /wherever/you/want/to/clone && \
  --mount=type=ssh git clone [email protected]:your-project.git
  1. And now you can finally build your Dockerfile (with buildkit enabled)
DOCKER_BUILDKIT=1 docker build . --ssh default

As you cannot currently pass console parameters to build in docker-compose, this solution is not available yet for docker-compose, but it should be soon (it's been done on github and proposed as a merge request)

Upvotes: 14

mosaad
mosaad

Reputation: 2381

For other people searching I had the same issue adding --ssh default flag made it work

Upvotes: 0

jaker
jaker

Reputation: 7840

Another option is to use a multi-stage docker build to ensure that your SSH keys are not included in the final image.

As described in my post you can prepare your intermediate image with the required dependencies to git clone and then COPY the required files into your final image.

Additionally if we LABEL our intermediate layers, we can even delete them from the machine when finished.

# Choose and name our temporary image.
FROM alpine as intermediate
# Add metadata identifying these images as our build containers (this will be useful later!)
LABEL stage=intermediate

# Take an SSH key as a build argument.
ARG SSH_KEY

# Install dependencies required to git clone.
RUN apk update && \
    apk add --update git && \
    apk add --update openssh

# 1. Create the SSH directory.
# 2. Populate the private key file.
# 3. Set the required permissions.
# 4. Add github to our list of known hosts for ssh.
RUN mkdir -p /root/.ssh/ && \
    echo "$SSH_KEY" > /root/.ssh/id_rsa && \
    chmod -R 600 /root/.ssh/ && \
    ssh-keyscan -t rsa github.com >> ~/.ssh/known_hosts

# Clone a repository (my website in this case)
RUN git clone [email protected]:janakerman/janakerman.git

# Choose the base image for our final image
FROM alpine

# Copy across the files from our `intermediate` container
RUN mkdir files
COPY --from=intermediate /janakerman/README.md /files/README.md

We can then build:

MY_KEY=$(cat ~/.ssh/id_rsa)
docker build --build-arg SSH_KEY="$MY_KEY" --tag clone-example .

Prove our SSH keys are gone:

docker run -ti --rm clone-example cat /root/.ssh/id_rsa

Clean intermediate images from the build machine:

docker rmi -f $(docker images -q --filter label=stage=intermediate)

Upvotes: 39

crooksey
crooksey

Reputation: 8809

My key was password protected which was causing the problem, a working file is now listed below (for help of future googlers)

FROM ubuntu

MAINTAINER Luke Crooks "[email protected]"

# Update aptitude with new repo
RUN apt-get update

# Install software 
RUN apt-get install -y git
# Make ssh dir
RUN mkdir /root/.ssh/

# Copy over private key, and set permissions
# Warning! Anyone who gets their hands on this image will be able
# to retrieve this private key file from the corresponding image layer
ADD id_rsa /root/.ssh/id_rsa

# Create known_hosts
RUN touch /root/.ssh/known_hosts
# Add bitbuckets key
RUN ssh-keyscan bitbucket.org >> /root/.ssh/known_hosts

# Clone the conf files into the docker container
RUN git clone [email protected]:User/repo.git

Upvotes: 379

Nomce
Nomce

Reputation: 775

For bitbucket repository, generate App Password (Bitbucket settings -> Access Management -> App Password, see the image) with read access to the repo and project.

bitbucket user menu

Then the command that you should use is:

git clone https://username:[email protected]/reponame/projectname.git

Upvotes: 23

Calvin Froedge
Calvin Froedge

Reputation: 16373

There's no need to fiddle around with ssh configurations. Use a configuration file (not a Dockerfile) that contains environment variables, and have a shell script update your docker file at runtime. You keep tokens out of your Dockerfiles and you can clone over https (no need to generate or pass around ssh keys).

Go to Settings > Personal Access Tokens

  • Generate a personal access token with repo scope enabled.
  • Clone like this: git clone https://[email protected]/user-or-org/repo

Some commenters have noted that if you use a shared Dockerfile, this could expose your access key to other people on your project. While this may or may not be a concern for your specific use case, here are some ways you can deal with that:

  • Use a shell script to accept arguments which could contain your key as a variable. Replace a variable in your Dockerfile with sed or similar, i.e. calling the script with sh rundocker.sh MYTOKEN=foo which would replace on https://{{MY_TOKEN}}@github.com/user-or-org/repo. Note that you could also use a configuration file (in .yml or whatever format you want) to do the same thing but with environment variables.
  • Create a github user (and generate an access token for) for that project only

Upvotes: 105

Marcin R
Marcin R

Reputation: 1601

You should create new SSH key set for that Docker image, as you probably don't want to embed there your own private key. To make it work, you'll have to add that key to deployment keys in your git repository. Here's complete recipe:

  1. Generate ssh keys with ssh-keygen -q -t rsa -N '' -f repo-key which will give you repo-key and repo-key.pub files.

  2. Add repo-key.pub to your repository deployment keys.
    On GitHub, go to [your repository] -> Settings -> Deploy keys

  3. Add something like this to your Dockerfile: 

    ADD repo-key /
RUN \
  chmod 600 /repo-key && \  
  echo "IdentityFile /repo-key" >> /etc/ssh/ssh_config && \  
  echo -e "StrictHostKeyChecking no" >> /etc/ssh/ssh_config && \  
  // your git clone commands here...

Note that above switches off StrictHostKeyChecking, so you don't need .ssh/known_hosts. Although I probably like more the solution with ssh-keyscan in one of the answers above.

Upvotes: 127

Related Questions