4.6 Git on the Server - Smart HTTP

We now have authenticated access through SSH and unauthenticated access through git://, but there is also a protocol that can do both at the same time. Setting up Smart HTTP is basically just enabling a CGI script that is provided with Git called git-http-backend on the server. This CGI will read the path and headers sent by a git fetch or git push to an HTTP URL and determine if the client can communicate over HTTP (which is true for any client since version 1.6.6). If the CGI sees that the client is smart, it will communicate smartly with it, otherwise it will fall back to the dumb behavior (so it is backward compatible for reads with older clients).

Let’s walk through a very basic setup. We’ll set this up with Apache as the CGI server. If you don’t have Apache setup, you can do so on a Linux box with something like this:

This also enables the mod_cgi, mod_alias, mod_env, and mod_rewrite modules, which are all needed for this to work properly.

You’ll also need to set the Unix user group of the /srv/git directories to www-data so your web server can read- and write-access the repositories, because the Apache instance running the CGI script will (by default) be running as that user:

Next we need to add some things to the Apache configuration to run the git-http-backend as the handler for anything coming into the /git path of your web server.

If you leave out GIT_HTTP_EXPORT_ALL environment variable, then Git will only serve to unauthenticated clients the repositories with the git-daemon-export-ok file in them, just like the Git daemon did.

Finally you’ll want to tell Apache to allow requests to git-http-backend and make writes be authenticated somehow, possibly with an Auth block like this:

That will require you to create a .htpasswd file containing the passwords of all the valid users. Here is an example of adding a “schacon” user to the file:

There are tons of ways to have Apache authenticate users, you’ll have to choose and implement one of them. This is just the simplest example we could come up with. You’ll also almost certainly want to set this up over SSL so all this data is encrypted.

We don’t want to go too far down the rabbit hole of Apache configuration specifics, since you could well be using a different server or have different authentication needs. The idea is that Git comes with a CGI called git-http-backend that when invoked will do all the negotiation to send and receive data over HTTP. It does not implement any authentication itself, but that can easily be controlled at the layer of the web server that invokes it. You can do this with nearly any CGI-capable web server, so go with the one that you know best.