Git Book - Transfer Protocols

Transfer Protocols

Here we will go over how clients and servers talk to each other to transfer Git data around.

Fetching Data over HTTP

Fetching over an http/s URL will make Git use a slightly dumber protocol. In this case, all of the logic is entirely on the client side. The server requires no special setup - any static webserver will work fine if the git directory you are fetching from is in the webserver path.

In order for this to work, you do need to run a single command on the server repo everytime anything is updated, though - git update-server-info, which updates the objects/info/packs and info/refs files to list which refs and packfiles are available, since you can't do a listing over http. When that command is run, the objects/info/packs file looks something like this:

P pack-ce2bd34abc3d8ebc5922dc81b2e1f30bf17c10cc.pack
P pack-7ad5f5d05f5e20025898c95296fe4b9c861246d8.pack

So that if the fetch can't find a loose file, it can try these packfiles. The info/refs file will look something like this:

184063c9b594f8968d61a686b2f6052779551613    refs/heads/development
32aae7aef7a412d62192f710f2130302997ec883    refs/heads/master

Then when you fetch from this repo, it will start with these refs and walk the commit objects until the client has all the objects that it needs.

For instance, if you ask to fetch the master branch, it will see that master is pointing to 32aae7ae and that your master is pointing to ab04d88, so you need 32aae7ae. You fetch that object

CONNECT http://myserver.com
GET /git/myproject.git/objects/32/aae7aef7a412d62192f710f2130302997ec883 - 200

and it looks like this:

tree aa176fb83a47d00386be237b450fb9dfb5be251a
parent bd71cad2d597d0f1827d4a3f67bb96a646f02889
author Scott Chacon <schacon@gmail.com> 1220463037 -0700
committer Scott Chacon <schacon@gmail.com> 1220463037 -0700

added chapters on private repo setup, scm migration, raw git

So now it fetches the tree aa176fb8:

GET /git/myproject.git/objects/aa/176fb83a47d00386be237b450fb9dfb5be251a - 200

which looks like this:

100644 blob 6ff87c4664981e4397625791c8ea3bbb5f2279a3    COPYING
100644 blob 97b51a6d3685b093cfb345c9e79516e5099a13fb    README
100644 blob 9d1b23b8660817e4a74006f15fae86e2a508c573    Rakefile

So then it fetches those objects:

GET /git/myproject.git/objects/6f/f87c4664981e4397625791c8ea3bbb5f2279a3 - 200
GET /git/myproject.git/objects/97/b51a6d3685b093cfb345c9e79516e5099a13fb - 200
GET /git/myproject.git/objects/9d/1b23b8660817e4a74006f15fae86e2a508c573 - 200

It actually does this with Curl, and can open up multiple parallel threads to speed up this process. When it's done recursing the tree pointed to by the commit, it fetches the next parent.

GET /git/myproject.git/objects/bd/71cad2d597d0f1827d4a3f67bb96a646f02889 - 200

Now in this case, the commit that comes back looks like this:

tree b4cc00cf8546edd4fcf29defc3aec14de53e6cf8
parent ab04d884140f7b0cf8bbf86d6883869f16a46f65
author Scott Chacon <schacon@gmail.com> 1220421161 -0700
committer Scott Chacon <schacon@gmail.com> 1220421161 -0700

added chapters on the packfile and how git stores objects

and we can see that the parent, ab04d88 is where our master branch is currently pointing. So, we recursively fetch this tree and then stop, since we know we have everything before this point. You can force Git to double check that we have everything with the '--recover' option. See git http-fetch for more information.

If one of the loose object fetches fails, Git will download the packfile indexes looking for the sha that it needs, then download that packfile.

It is important if you are running a git server that serves repos this way to implement a post-receive hook that runs the 'git update-server-info' command each time or there will be confusion.

Fetching Data with Upload Pack

For the smarter protocols, fetching objects is much more efficient. A socket is opened, either over ssh or over port 9418 (in the case of the git:// protocol), and the git fetch-pack command on the client begins communicating with a forked git upload-pack process on the server.

Then the server will tell the client which SHAs it has for each ref, and the client figures out what it needs and responds with a list of SHAs it wants and already has.

At this point, the server will generate a packfile with all the objects that the client needs and begin streaming it down to the client.

Let's look at an example.

The client connects and sends the request header. The clone command

$ git clone git://myserver.com/project.git

produces the following request:

0032git-upload-pack /project.git\000host=myserver.com\000

The first four bytes contain the hex length of the line (including 4 byte line length and trailing newline if present). Following are the command and arguments. This is followed by a null byte and then the host information. The request is terminated by a null byte.

The request is processed and turned into a call to git-upload-pack:

$ git-upload-pack /path/to/repos/project.git

This immediately returns information of the repo:

007c74730d410fcb6603ace96f1dc55ea6196122532d HEAD\000multi_ack thin-pack side-band side-band-64k ofs-delta shallow no-progress
003e7d1665144a3a975c05f1f43902ddaf084e784dbe refs/heads/debug
003d5a3f6be755bbb7deae50065988cbfa1ffa9ab68a refs/heads/dist
003e7e47fe2bd8d01d481f44d7af0531bd93d3b21c01 refs/heads/local
003f74730d410fcb6603ace96f1dc55ea6196122532d refs/heads/master
0000

Each line starts with a four byte line length declaration in hex. The section is terminated by a line length declaration of 0000.

This is sent back to the client verbatim. The client responds with another request:

0054want 74730d410fcb6603ace96f1dc55ea6196122532d multi_ack side-band-64k ofs-delta
0032want 7d1665144a3a975c05f1f43902ddaf084e784dbe
0032want 5a3f6be755bbb7deae50065988cbfa1ffa9ab68a
0032want 7e47fe2bd8d01d481f44d7af0531bd93d3b21c01
0032want 74730d410fcb6603ace96f1dc55ea6196122532d
00000009done

The is sent to the open git-upload-pack process which then streams out the final response:

"0008NAK\n"
"0023\002Counting objects: 2797, done.\n"
"002b\002Compressing objects:   0% (1/1177)   \r"
"002c\002Compressing objects:   1% (12/1177)   \r"
"002c\002Compressing objects:   2% (24/1177)   \r"
"002c\002Compressing objects:   3% (36/1177)   \r"
"002c\002Compressing objects:   4% (48/1177)   \r"
"002c\002Compressing objects:   5% (59/1177)   \r"
"002c\002Compressing objects:   6% (71/1177)   \r"
"0053\002Compressing objects:   7% (83/1177)   \rCompressing objects:   8% (95/1177)   \r"
...
"005b\002Compressing objects: 100% (1177/1177)   \rCompressing objects: 100% (1177/1177), done.\n"
"2004\001PACK\000\000\000\002\000\000\n\355\225\017x\234\235\216K\n\302"...
"2005\001\360\204{\225\376\330\345]z2673"...
...
"0037\002Total 2797 (delta 1799), reused 2360 (delta 1529)\n"
...
"<\276\255L\273s\005\001w0006\001[0000"

See the Packfile chapter previously for the actual format of the packfile data in the response.

Pushing Data

Pushing data over the git and ssh protocols is similar, but simpler. Basically what happens is the client requests a receive-pack instance, which is started up if the client has access, then the server returns all the ref head shas it has again and the client generates a packfile of everything the server needs (generally only if what is on the server is a direct ancestor of what it is pushing) and sends that packfile upstream, where the server either stores it on disk and builds an index for it, or unpacks it (if there aren't many objects in it)

This entire process is accomplished through the git send-pack command on the client, which is invoked by git push and the git receive-pack command on the server side, which is invoked by the ssh connect process or git daemon (if it's an open push server).