Commit 1e911e05 authored by Achilleas Pipinellis's avatar Achilleas Pipinellis Committed by Kamil Trzcinski

Refactor new CI build permissions model docs

parent 77271388
## Dependent projects # New CI build permission model
> Introduced in GitLab 8.12. > Introduced in GitLab 8.12.
GitLab 8.12 introduces a new [Build permissions model](../../user/permissions.md#builds-permissions). GitLab 8.12 has a completely redesigned build permission system. You can find
all discussion and all our concerns when choosing the current approach in issue
[#18994](https://gitlab.com/gitlab-org/gitlab-ce/issues/18994).
This opens an easy to use a way to access all dependent source codes: ---
1. Access project's `submodule`,
1. Access private Container Images,
1. Access project's and submodule LFS objects.
### Submodules Builds permissions should be tightly integrated with the permissions of a user
who is triggering a build.
> It often happens that while working on one project, you need to use another project from within it. The reasons to do it like that are:
> Perhaps it’s a library that a third party developed or that you’re developing separately and using in multiple parent projects.
> A common issue arises in these scenarios: you want to be able to treat the two projects as separate yet still be able to use one from within the other. - We already have a permissions system in place: group and project membership
> (from https://git-scm.com/book/en/v2/Git-Tools-Submodules) of users.
- We already fully know who is triggering a build (using `git push`, using the
web UI, executing triggers).
- We already know what user is allowed to do.
- We use the user permissions for builds that are triggered by the user.
- It opens a lot of possibilities to further enforce user permissions, like
allowing only specific users to access runners or use secure variables and
environments.
- It is simple and convenient that your build can access everything that you
as a user have access to.
- Short living unique tokens are now used, granting access for time of the build
and maximizing security.
With the new behavior, any build that is triggered by the user, is also marked
with their permissions. When a user does a `git push` or changes files through
the web UI, a new pipeline will be usually created. This pipeline will be marked
as created be the pusher (local push or via the UI) and any build created in this
pipeline will have the permissions of the pusher.
This allows us to make it really easy to evaluate the access for all projects
that have Git submodules or are using container images that the pusher would
have access too. **The permission is granted only for time that build is running.
The access is revoked after the build is finished.**
## Types of users
It is important to note that we have a few types of users:
- **Administrators**: CI builds created by Administrators will not have access
to all GitLab projects, but only to projects and container images of projects
that the administrator is a member of.That means that if a project is either
public or internal users have access anyway, but if a project is private, the
Administrator will have to be a member of it in order to have access to it
via another project's build.
- **External users**: CI builds created by external users will have access only
to projects to which user has at least reporter access. This rules out
accessing all internal projects by default,
This allows us to make the CI and permission system more trustworthy.
Let's consider the following scenario:
1. You are an employee of a company. Your company has a number of internal tools
hosted in private repositories and you have multiple CI builds that make use
of these repositories.
2. You invite a new user, a visitor, the external user. CI builds created by
that user do not have access to internal repositories, because the user also
doesn't have the access from within GitLab. You as an employee have to grant
explicit access for this user. This allows us to prevent from accidental data
leakage.
## Build token
A unique build token is generated for each project and it allows the user to
access all projects that would be normally accessible to the user creating that
build.
We try to make sure that this token doesn't leak by:
1. Securing all API endpoints to not expose the build token.
1. Masking the build token from build logs.
1. Allowing to use the build token **only** when build is running.
However, this brings a question about the Runners security. To make sure that
this token doesn't leak, you should also make sure that you configure
your Runners in the most possible secure way, by avoiding the following:
1. Any usage of Docker's `privileged` mode is risky if the machines are re-used.
1. Using the `shell` executor since builds run on the same machine.
By using an insecure GitLab Runner configuration, you allow the rogue developers
to steal the tokens of other builds.
## Debugging problems
With the new permission model in place, there may be times that your build will
fail. This is most likely because your project tries to access other project's
sources, and you don't have the appropriate permissions. In the build log look
for information about 403 or forbidden access messages
As an Administrator, you can verify that the user is a member of the group or
project they're trying to have access to, and you can impersonate the user to
retry the failing build in order to verify that everything is correct.
## Before GitLab 8.12
In versions before GitLab 8.12, all CI builds would use the CI Runner's token
to checkout project sources.
The project's Runner's token was a token that you could find under the
project's **Settings > CI/CD Pipelines** and was limited to access only that
project.
The project's Runner's token was used for registering new specific Runners
assigned to project and to checkout project sources.
The project Runners token could also be used with the GitLab Container Registry
for that project, allowing pulling and pushing Docker images from within the
CI build.
GitLab would create a special checkout URL like:
```
https://gitlab-ci-token:<project-runners-token>/gitlab.com/gitlab-org/gitlab-ce.git
```
And then the users could also use it in their CI builds all Docker related
commands to interact with GitLab Container Registry. For example:
```
docker login -u gitlab-ci-token -p $CI_BUILD_TOKEN registry.gitlab.com
```
Using single token had multiple security implications
- The token would be readable to anyone who had developer access to a project
that could run CI builds, allowing the developer to register any specific
Runner for that project.
- The token would allow to access only the project's sources, forbidding from
accessing any other projects.
- The token was not expiring and was multi-purpose: used for checking out sources,
for registering specific runners and for accessing a project's container
registry with read-write permissions.
All the above led to a new permission model for builds that was introduced
with GitLab 8.12.
## Making use of the new CI build permissions model
With the new build permission model, there is now an easy way to access all
dependent source code in a project. That way, we can:
1. Access a project's Git submodules
1. Access private container images
1. Access project's and submodule LFS objects
Let's see how that works with Git submodules and private Docker images hosted on
the container registry.
## Git submodules
>
It often happens that while working on one project, you need to use another
project from within it. Perhaps it’s a library that a third party developed or
you’re developing a project separately and are using it in multiple parent
projects.
A common issue arises in these scenarios: you want to be able to treat the two
projects as separate yet still be able to use one from within the other.
>
_Excerpt from the [Git website][git-scm] about submodules._
If dealing with submodules, your project will probably have a file named
`.gitmodules`. And this is how it usually looks like:
Your project usually have a file named `.gitmodules`.
This file usually looks like that:
``` ```
[submodule "tools"] [submodule "tools"]
path = tools path = tools
url = git@gitlab.com/group/tools.git url = git@gitlab.com/group/tools.git
``` ```
Before 8.12 you had to do a multiple workarounds (ex. [SSH keys](../ssh_keys/README.md)) > **Note:**
in order to access the sources of `gitlab.com/group/tools`. If you are not using GitLab 8.12 or higher, you would need to work your way
around this issue in order to access the sources of `gitlab.com/group/tools`
GitLab 8.12 uses your permissions to evaluate what a CI build can access. (e.g., use [SSH keys](../ssh_keys/README.md)).
More information about how this system works can be found here: [Build permissions model](../../user/permissions.md#builds-permissions). >
With GitLab 8.12 onward, your permissions are used to evaluate what a CI build
can access. More information about how this system works can be found in the
[Build permissions model](../../user/permissions.md#builds-permissions).
To make use of a new changes you have to update your `.gitmodules` file to use a relative URL. To make use of the new changes you have to update your `.gitmodules` file to
use a relative URL.
Let's consider the following example: Let's consider the following example:
1. Your project is located at https://gitlab.com/secret-group/my-project, 1. Your project is located at `https://gitlab.com/secret-group/my-project`.
1. To checkout your sources you usually use a SSH address: `git@gitlab.com:secret-group/my-project.git`, 1. To checkout your sources you usually use an SSH address like
1. Your project depend on https://gitlab.com/group/tools, `git@gitlab.com:secret-group/my-project.git`.
1. Your project depends on `https://gitlab.com/group/tools`.
1. You have the `.gitmodules` file with above content. 1. You have the `.gitmodules` file with above content.
Since that you can use a relative URLs for your `.gitmodules` configuration Since Git allows the usage of relative URLs for your `.gitmodules` configuration,
it easily allows you to use an HTTP cloning for all your CI builds, this easily allows you to use HTTP for cloning all your CI builds and SSH
and SSH clonning for all your local checkouts. for all your local checkouts.
If you change the `url` of your `tools` dependency: For example, if you change the `url` of your `tools` dependency, from
``` `git@gitlab.com/group/tools.git` to `../../group/tools.git`, this will instruct
git@gitlab.com/group/tools.git => ../../group/tools.git Git to automatically deduce the URL that should be used when cloning sources.
``` Whether you use HTTP or SSH, Git will use that same channel and it will allow
to make all your CI builds use HTTPS (because GitLab CI uses HTTPS for cloning
your sources), and all your local clones will continue using SSH.
It will instruct GIT to automatically deduce a URL that should be used when cloning sources. Given the above explanation, your `.gitmodules` file should eventually look
Whether you used a HTTP or SSH it will instruct GIT to use the same channel. like this:
And this will allow to make all your CI builds to use HTTPS (because GitLab CI uses HTTPS for cloning your sources),
and all your local clones will continue using SSH.
Given the above explanation, your `.gitmodules` file should look like this:
``` ```
[submodule "tools"] [submodule "tools"]
path = tools path = tools
url = ../../group/tools.git url = ../../group/tools.git
``` ```
However, you have to instruct GitLab CI to clone your submodules as this is not done automatically. However, you have to explicitly tell GitLab CI to clone your submodules as this
You can achieve that by adding a `before_script` section to your `.gitlab-ci.yml` with `git submodule` command: is not done automatically. You can achieve that by adding a `before_script`
section to your `.gitlab-ci.yml`:
``` ```
before_script: before_script:
...@@ -72,10 +228,12 @@ test: ...@@ -72,10 +228,12 @@ test:
- run-my-tests - run-my-tests
``` ```
This will make GitLab CI to initialize (fetch) and update (checkout) all your submodules recursively. This will make GitLab CI initialize (fetch) and update (checkout) all your
submodules recursively.
It can happen that your environment or your Docker Image does not have a git installed. In case your environment or your Docker image doesn't have Git installed,
You have to either ask your Administrator or install the missing dependency yourself: you have to either ask your Administrator or install the missing dependency
yourself:
``` ```
# Debian / Ubuntu # Debian / Ubuntu
...@@ -97,17 +255,21 @@ before_script: ...@@ -97,17 +255,21 @@ before_script:
### Container Registry ### Container Registry
With the update permission model we also extended support for accessing Container Registries for private projects. With the update permission model we also extended the support for accessing
Container Registries for private projects.
> **Note:**
As GitLab Runner 1.6 doesn't yet incorporate the introduced changes for
permissions, this makes the `image:` directive to not work with private projects
automatically. The manual configuration by an Administrator is required to use
private images. We plan to remove that limitation in one of the upcoming releases.
> Note: As of 1.6 the GitLab Runner doesn't yet incorporate the introduced changes for permissions. Your builds can access all container images that you would normally have access
> This makes a `image:` to not work with private projects automatically. to. The only implication is that you can push to the Container Registry of the
> The manual configuration by Administrator is required to use private images. project for which the build is triggered.
> We plan to remove that limitation in one of the upcoming releases.
Your builds can access all container images that you would normally have access to. This is how an example usage can look like:
The only implication is that you can push to Container Registry of project for which the build is triggered.
This is how the example usage can look like:
``` ```
test: test:
script: script:
...@@ -115,3 +277,5 @@ test: ...@@ -115,3 +277,5 @@ test:
- docker pull $CI_REGISTRY/group/other-project:latest - docker pull $CI_REGISTRY/group/other-project:latest
- docker run $CI_REGISTRY/group/other-project:latest - docker run $CI_REGISTRY/group/other-project:latest
``` ```
[git-scm]: https://git-scm.com/book/en/v2/Git-Tools-Submodules
...@@ -147,49 +147,6 @@ GitLab 8.12 has completely redesigned build permission system. ...@@ -147,49 +147,6 @@ GitLab 8.12 has completely redesigned build permission system.
You can find all discussion and all our concerns when choosing the current approach: You can find all discussion and all our concerns when choosing the current approach:
https://gitlab.com/gitlab-org/gitlab-ce/issues/18994 https://gitlab.com/gitlab-org/gitlab-ce/issues/18994
We decided that builds permission should be tightly integrated with a permission
of a user who is triggering a build.
The reason to do it like that:
- We already have permission system in place: group and project membership of users,
- We already fully know who is triggering a build (using git push, using web, executing triggers),
- We already know what user is allowed to do,
- We use the user permission for builds that are triggered by him,
- This opens us a lot of possibilities to further enforce user permissions, like:
allowing only specific users to access runners, secure variables and environments,
- It is simple and convenient, that your build can access to everything that you have access to,
- We choose to short living unique tokens, granting access for time of the build,
Currently, any build that is triggered by the user, it's also signed with his permissions.
When user do `git push` or changes files through web (**the pusher**),
we will usually create a new Pipeline.
The Pipeline will be signed as created be the pusher.
Any build created in this pipeline will have the permissions of **the pusher**.
This allows us to make it really easy to evaluate access for all dependent projects,
container images that the pusher would have access too.
The permission is granted only for time that build is running.
The access is revoked after the build is finished.
It is important to note that we have a few types of Users:
- Administrators: CI builds created by Administrators would not have access to all GitLab projects,
but only to projects and container images of projects that the user is a member of or that are either public, or internal,
- External users: CI builds created by external users will have access only to projects to which user has at least reporter access,
this rules out accessing all internal projects by default,
This allows us to make the CI and permission system more trustable.
Let's consider the following scenario:
1. You are an employee of the company. Your company have number of internal tool repositories.
You have multiple CI builds that make use of this repositories.
2. You invite a new user, a visitor, the external user. CI builds created by that user do not have access to internal repositories,
because user also doesn't have the access from within GitLab. You as an employee have to grant explicit access for this user.
This allows us to prevent from accidental data leakage.
### Build privileges ### Build privileges
This table shows granted privileges for builds triggered by specific types of users: This table shows granted privileges for builds triggered by specific types of users:
...@@ -212,71 +169,3 @@ This table shows granted privileges for builds triggered by specific types of us ...@@ -212,71 +169,3 @@ This table shows granted privileges for builds triggered by specific types of us
[^3]: Only if user is not external one. [^3]: Only if user is not external one.
[^4]: Only if user is a member of the project. [^4]: Only if user is a member of the project.
### Build token
The above gives a question about trustability of build token.
Unique build token is generated for each project.
This build token allows to access all projects that would be normally accessible
to the user creating that build.
We try to make sure that this token doesn't leak.
We do that by:
1. Securing all API endpoints to not expose the build token,
1. Masking the build token from build logs,
1. Allowing to use the build token only when build is running,
However, this brings a question about runners security.
To make sure that this token doesn't leak you also make sure that you configure
your runners in most secure possible way, by avoiding using this configurations:
1. Any usage of `privileged` mode if the machines are re-used is risky,
1. Using `shell` executor,
By using in-secure GitLab Runner configuration you allow the rogue developers
to steal the tokens of other builds.
### Debugging problems
It can happen that some of the users will complain that CI builds do fail for them.
It is most likely that your project access other projects sources,
and the user doesn't have the permissions.
In the build log look for information about 403 or forbidden access.
You then as Administrator can verify that the user is a member of the group or project,
and you when impersonated as the user can retry a failing build
on behalf of the user to verify that everything is correct.
### Before 8.12
In versions before 8.12 all CI builds would use runners token to checkout project sources.
The project runners token was a token that you would find in
[CI/CD Pipelines](https://gitlab.com/my-group/my-project/pipelines/settings).
The project runners token was used for registering new specific runners assigned to project
and to checkout project sources.
The project runners token could also be used to use GitLab Container Registry for that project,
allowing to pull and push Docker images from within CI build.
This token was limited to access only that project.
GitLab would create an special checkout URL:
```
https://gitlab-ci-token:<project-runners-token>/gitlab.com/gitlab-org/gitlab-ce.git
```
User could also use in his CI builds all docker related commands
to interact with GitLab Container Registry:
```
docker login -u gitlab-ci-token -p $CI_BUILD_TOKEN registry.gitlab.com
```
Using single token had multiple security implications:
- Token would be readable to anyone who has developer access to project who could run CI builds,
allowing to register any specific runner for a project,
- Token would allow to access only project sources,
forbidding to accessing any other projects,
- Token was not expiring, and multi-purpose: used for checking out sources,
for registering specific runners and for accessing project's container registry with read-write permissions
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment