remove non-tsa content authored by anarcat's avatar anarcat
# git.torproject.org and associated services
[[_TOC_]]
## Overview
Our git setup consists of three interdependent services:
* git-rw.torproject.org: ssh accessible and writeable git repositories
* git.torproject.org: https:// read-only anonymous access
* [gitweb.torproject.org](https://gitweb.torproject.org/): web browsing repositories
When a developer pushes to git-rw, the repository is mirrored to git and so
made available via the gitweb service.
## Regular repositories
### Creating a new repository
Creating a new top-level repository is not something that should be done often.
The top-level repositories are all shown on the gitweb, and we'd like to keep
the noise down. If you're not sure if you need a top-level repository then
perhaps request a user repository first, and use that until you know you need
a top-level repository.
Some projects, for example pluggable-transports, have a path hierachy for their
repositories. This should be encouraged to help keep this organised.
A request for a new top-level repository should include: the users that should
have access to it, the repository name (including any folder it should live in),
and a short description. If the users that should have access to this repository
should be kept in sync with some other repository, a group might be created or
re-used as part of the request.
For example:
```
Please create a new repository metrics/awesome-pipeline.git.
This should be accessible by the same set of users that have access to the
metrics-cloud repository.
The description for the repository is: Tor Metrics awesome pipeline repository.
This message was signed for trac.torproject.org on 2018-10-16 at 19:00:00 UTC.
```
The git team may ask for additional information to clarify the request if
necessary, and may ask for replies to that information to be signed if they
would affect the access to the repository. In the case that replies are to be
signed, include the ticket number in the signed text to avoid replay attacks.
The git team member will edit the gitolite configuration to add a new block
(alphabetically sorted within the configuration file) that looks like the
following:
```
repo metrics-cloud
RW = @metrics-cloud
config hooks.email-enabled = true
config hooks.mailinglist = tor-commits@lists.torproject.org
config hooks.irc-enabled = true
config hooks.ircproject = or
config hooks.githuburl = torproject/metrics-cloud
config hooks.gitlaburl = torproject/metrics/metrics-cloud
metrics-cloud "The Tor Project" = "Configurations for Tor Metrics cloud orchestration"
```
Deconstructing this:
```
repo metrics-cloud
```
Starts a repository block.
```
RW = @metrics-cloud
```
Allows non-destructive read/write but not branch/tag deletion or
non-fast-forward pushes. Alternatives would include "R" for read-only, or
"RW+" to allow for destructive actions. We only allow destructive actions for
user's personal repositories.
In this case, the permissions are delegated to a group (starting with @) and
not an individual user.
```
config hooks.email-enabled = true
config hooks.mailinglist = tor-commits@lists.torproject.org
```
This enables the email hook to send one email per commit to the commits list.
For all top-level repositories, the mailing list should be
tor-commits@lists.torproject.org.
```
config hooks.irc-enabled = true
config hooks.ircproject = or
```
This enables the IRC hook to send one message per commit to an IRC channel. If
the project is set to "or" the messages will be sent to #tor-bots.
```
config hooks.githuburl = torproject/metrics-cloud
config hooks.gitlaburl = torproject/metrics/metrics-cloud
```
These enable pushing a mirror to external services. The external service will
have to be configured to accept these pushes, and we should avoid adding
mirror URLs where things aren't configured yet so we don't trigger any IPS or
abuse detection system by making loads of bad push attempts.
```
metrics-cloud "The Tor Project" = "Configurations for Tor Metrics cloud orchestration"
```
The last line of this file is what is used to provide configuration to gitweb.
Starting with the path, then the owner, then the short description.
Upon push, the new repository will be created. It may take some minutes to
appear on the gitweb. Do not fear, the old list that did not yet include the
new repository has just been cached.
Push takes ages. Don't Ctrl-C it or you can end up in an inconsistent state.
Just let it run. A future git team member might work on backgrounding the
sync task.
Groups are defined at the top of the file, again in alphabetical order
(not part of the repository block):
```
@metrics-cloud = karsten irl
```
### Adding developers to a repository
If you want access to an existing repository please have somebody who already
has access to ask that you be added by filing a trac ticket. This should be GPG
signed as above.
[Request a user be added to an existing repository](https://trac.torproject.org/projects/tor/newticket?summary=Please%20add%20%3Cusername%3E%20to%20repository%20%3Crepository%3E.git&component=Internal%20Services/Service%20-%20git&type=task)
The git team member will either add a permissions line to the configuration for
the repository or will add a username to the group, depending on how the
repository is configured.
### Deleting accidentally pushed tags/branches
These requests are for a destructive action and should be signed. You should
also sanity check the request and not just blindly copy/paste the list of
branch names.
The git team member will need to:
1. Edit the gitolite configuration to allow RW+ access for the specified branch
or tag.
2. Push an empty reference to the remote reference to delete it. In doing this,
all the hooks will run ensuring that the gitweb mirror and all other external
mirrors are kept in sync.
3. Revert the commit that gave the git team member this access.
The additional permission line will look something like:
```
RW+ refs/heads/travis-ci = irl
RW+ refs/tags/badtag-v1.0 = irl
```
This is to protect the git team member from accidentally deleting everything,
do not just give yourself `RW+` permissions for the whole repository unless you
are feeling brave, even when someone has accidentally pushed their entire
history of personal branches to the canonical repository.
## User repositories
Developers who have a tpo LDAP account can request personal git repositories be
created on our git infrastructure. Please file a ticket in Trac using the link
below. User repositories have the path `user/<username>/<repository>.git`.
[Request a new user repository](https://trac.torproject.org/projects/tor/newticket?summary=Please%20create%20new%20repository%20user/%3Cusername%3E/%3Crepository%3E.git&component=Internal%20Services/Service%20-%20git&type=task)
This request should contain: username, repository name, and a short
description. Here is an example where irl is requesting a new example
repository:
```
Please create a new user repository user/irl/example.git.
The description for the repository is: Iain's example repository.
This message was signed for trac.torproject.org on 2018-10-16 at 19:00:00 UTC.
```
Please use GPG to clearsign this text, it will be checked against the GPG key
that you have linked to you in our LDAP. Additionally, ensure that it is
wrapped as a code block (within !{{{ }}}).
There have not yet been any cases where user repositories have allowed access
by other users than the owner. Let's keep it that way or this will get
complicated.
Users will have full access to their own repos and can therefore delete
branches, tags, and perform non-fast-forward pushes.
## Learning what git repos you can read/write
Once you have an LDAP account and have an ssh key set up for it, run:
```
ssh git@git-rw.torproject.org
```
and it will tell you what bits you have on which repos. The first column is who
can read (@ for everybody, R for you, blank for not you), and the second column
is who can write (@ for everybody, W for you, blank for not you).
## Commit hooks
There are a variety of commit hooks that are easy to add for your git repo,
ranging from irc notifications to email notifications to github auto-syncing.
Clone the gitolite-admin repo and look at the "config hooks" lines for
examples. You can request changes by filing a trac ticket as described above,
or just request the hooks when you first ask for your repo to be set up.
Hooks are stored in `/srv/git.torproject.org/git-helpers` on the
server.
### Standard Commit Hooks for Canonical Repositories
Changes to most repositories are reported to:
* the #tor-bots IRC channel (or #tor-internal for private admin repositories)
* Some repositories have a dedicated mailing list for commits at https://lists.torproject.org
### GitHub and GitLab Mirrors
Some repositories are mirrored to https://github.com/torproject
organization and to the https://gitlab.torproject.org/ server, through
gitlite hooks.
## Admin details
git-rw runs on `cupani.torproject.org` and runs as the git user. Users in the
gitolite (gid 1504) group can become the git user. The gitolite installation
is contained inside `/srv/git.torproject.org` with the repositories being found
in the `repositories` folder there.
The gitolite installation itself is *not* from Debian packages. It's a
manual install, in `/srv/git.torproject.org/gitolite/src`, of an
extremely old version (`v0.95-38-gb0ce84d`, december 2009).
Anonymous git and gitweb run on `vineale.torproject.org` and as the gitweb
user. Users in the gitweb (gid 1505) group can become the gitweb user.
Data for these services can be found in `/srv/gitweb.torproject.org`.
The gitolite configuration is found at
`git@git-rw.torproject.org:gitolite-admin.git` and is not mirrored to gitweb.
The `gitolite` group on the `git-rw` server defined in LDAP and has
total control of the gitolite installation, as its members can `sudo`
to git.
The `git` user gets redirected through the
`/srv/git.torproject.org/gitolite/src/gl-auth-command` through the
`/etc/ssh/userkeys/git` `authorized_keys` file. This, in turn, gets
generated from LDAP, somewhere inside the `ud-generate` command,
because `exportOptions` is set to `GITOLITE` on the `cupani` host. All
users with a valid LDAP account get their SSH key added to the list
and only gitolite configuration restricts further access.
Access to push to this repository is controlled by the
`gitolite-admin` repository entry in the gitolite configuration file,
and not by LDAP groups.
## Potential problems to solve
### gitweb out of sync
If vineale is down for an extended period of time, it's a good idea to trigger
a re-sync of all the repositories to ensure that the latest version is available
to clone from the anonymous endpoints.
Create an empty commit in the gitolite-admin.git repository using:
```
git commit -m "trigger resync" --allow-empty
```
and push this commit. This will run through the post-commit hook that includes
syncing everything.