GNU Linux (Debian and others) – how to create one’s own private git repository in under 10min (git over ssh)

25.Apr.2020

home of git’s git repo: https://git-scm.com/

# to clone git's git repo
git clone https://git.kernel.org/pub/scm/git/git.git
# as can be seen there, it is STILL under active development

written in primarily in C, with GUI and programming scripts written in Shell script, Perl, Tcl and Python^[4]^[5]
one of it’s credos is: “distributed-even-if-your-workflow-isnt“
git was initially written by Linus Torvalds on the idea: how to create source code management system that does not suck
- well was pretty happy with svn, but of course Mr Torvalds manages a software project (the GNU Linux kernel) with THOUSANDS of developers and MILLIONS of commits a day, so there was the plausible wish for something that was specific to that task, if it works for others, that might be a nice sideffect.
- currently maintained by Junio C Hamano. Numerous contributions have come from the Git mailing list git ÄT vger DOT kernel DOT org. openhub.net/p/git/contributors/summary
  - facesofopensource.com
    
    just btw: “Black Duck Open Hub, formerly Ohloh,^[2] is a website which provides a web services suite and online community platform that aims to index the open-source software development community. It was founded by former M$ managers Jason Allen and Scott Collison in 2004 and joined by the developer Robin Luckey” (src)
  - have a clone of git.git itself, the output of git-shortlog(1) and git-blame(1) can show you the authors for specific parts of the project.
source code management is very useful when the developer team is x > 1 developer.
it allows fast syncing of projects (e.g. from work-desktop to home-laptop and vice versa)
It seems to be a very holistic/complete/robust design and implemented file-based (no mysql server needed) source control system,
that has some complexity to it. Complexity might be reduced by using scripts or tools (vscode, eclipse) to simplify the tasks.
- with all complex systems: usually only those devs that wrote it TRULY understand why-what-is-going-on-right-or-wrong-under-the-hood
git is not only good for source code management
some even use it as content management system (version control of documents and other files (anything where organized cooperation between multiple users/developers/designers/content creators is important)).
GitHub is not run by Linux Foundation or Linus, it is basically not more than “web frontend” for a centralized source code on a not-owned-by-the-user-server system, sold for $7.5 billions to M$
- it is great for sharing open source src software
- but setting up the user-soon-dev’s own ssh vm (virtual machine also refered to as vps or “cloud”) debian 12 based server is fast and not-so-expensive: model: CAX11, vCPU 2x Ampere ARM + RAM 4 GB + Disk space 40 GB + Traffic 20 TB + IPv4 ~ 4.50 per month (2024-03)
but it is possible to setup a self-hosted gitlab (nice web gui)
git-server-side-storage-style: when trying to find the files pushed to the git-repo-server, by their filename, the user will probably not get lucky because…
- - “git logically stores each file under its SHA1 as filename, because: if the user has two files with exactly the same content in a repository, only one copy is stored” (src) (Torvalds, what have you done? X-D)

just because Torvalds likes it does not mean it’s the right thing for everyone and everything

the good about git over ssh:

+ free + open source
+ relatively fast and easy setup (scroll down): requirements: a working secure open ssh server
+ well thought out concept (time will tell)
+ high quality (almost-error-free) software (time will show)
+ well documented INCLUDING video tutorials (QUALITY UNTESTED)
~ various gui front ends SHOULD (if they themselves don’t get too complicated) make interaction with git repos easier but well
- - + one written in perl even hosted where the git src are hosted on kernel.org (the Linus approved gui?)
  - https://git-scm.com/docs/gitk
  - install like:
  - ```
  su - root
```
- ```
apt update; apt install git gitk git-gui;
# start it like
git gui
```
  - look like this:
there is more to be found here: “Cogito (originally git-pasky) is a revision control system layered on top of Git. It is historically the first Git frontend, which appeared in April 2005″ (src)

the problem with git over ssh:

– renaming-the-wheel: why rename something that already has a name? a commit is basically just an “upload” of changes of code, so call it an “upload” (it’s too late now)
- imho this is a mistake done often by various commercial companies for marketing purposes: coming up with “buzzwords” for things that already exist and have a name = causes inefficiencies = new devs have to find out that “what is what” and that a “commit” is just a “upload” of changes of code
– overhead of having to learn another complex software, this complexity can get in the way of actually coding
– if all devs push in one branch (which they should not do) it might be hard to revert changes, which is one of the most basic and fundamental functions of a code management system
– if passwordless public-private-ssh-key auth shall be used to commit changes, the private key of this user needs to be on every system that wants to clone a
repo, so if the private key get’s in the wrong hands, the whole git-ssh-server is in danger of becoming ranomeware-deleted
- — the result is: brute-forcable password auth is enabled, except to be brute-foced night-and-day
  - eclipse egit offers to save the password in a more-or-less secure way (expect it to be less secure)
– it happened more than once, that devs git-commit-uploaded sensitive information as access tokens, passwords and more-or-less private mail addresses to github
- – so while it is possible to set a filter file .gitignore to ignore certain files and folders it is often forgotten and then security problems can happen 🙁 (especially on repos that are publicly visible, not so much on the dev’s own ssh server… there bruteforce password attacks are the bigger threat)

be aware! git is a complex system!

only god & nobody is perfect thus complexity means: errors & problems and this means: frustrations

if the admin-dev-user does not plan to use it on a daily basis, it might not be worth diving down on it and maybe (for small teams might be okay) use (S)FTP? (also very easy to setup)

there definately need to be more tools to simplify the usage of git (pull, merge, push)

GNU Linux SFTP (ftp over ssh) – how to limit users to their home directories

git-over-ssh is a cool combination

every GNU Linux / OSX and even Windows (putty) can do ssh
ssh enables secure encrypted transfer of the files
ssh server is very easy to setup and with strong passwords also very secure (filtering of brute force traffic should non the less be done as well, bash savy admins can look here for inspiration)
does DELTA transfer meaning: small changes to big files can be transferred/synced fast (faster internet/more bandwidth is always a plus) (so in theory even large movie.mp4 and sound.wav files should be possible)

given that the requirements:

make sure password (or more convenient public-private key auth) ssh login from client->server works
it seems that as little as 256MByte of (VM asigned) RAM is sufficient
- so the cheapest vps from hetzner is waaaaay enough

free -m; # superbly efficient
              total        used        free      shared  buff/cache   available
Mem:            231          65          30           2         135         154
Swap:          1020           0        1020

are already in place (https://dwaves.de/2017/05/05/linux-ssh-generate-public-private-keys/)

one can setup one’s own git repository in 10min:

# tested on
hostnamectl; # client
   Static hostname: client
         Icon name: computer-desktop
           Chassis: desktop
  Operating System: Debian GNU/Linux 10 (buster)
            Kernel: Linux 4.19.0-8-amd64
      Architecture: x86-64

hostnamectl; # server
   Static hostname: git
         Icon name: computer-vm
           Chassis: vm
  Operating System: Debian GNU/Linux 10 (buster)
            Kernel: Linux 4.19.0-8-amd64
      Architecture: x86-64

# DEFINATELY enable ssh logging (not enabled per default on Debian)

# for conveniance an ll alias is recommended
# create ll alias to nicely display folder contents
alias ll='ls -lah --color --time-style=+%F'
# add this line to /etc/bashrc or /etc/bash.bashrc

# ===== on (ssh-enabled)server =====
# create repository directory (where to store project on server)
mkdir -p ~/projects/test.git
cd ~/projects/test.git

# initialize the server folder as git repository
git init --bare

# ===== on client =====
# create repository directory (where to store project on client)
mkdir -p ~/projects/test.git
cd ~/projects/test.git

# initialize the client folder as git repository
git init

# create some test data
echo -e "this is a test git repository created on $(date '+%Y-%m-%d-%H:%M:%S')" > README.txt;

# make git realize there are new files
git add .

# the git-system will not proceed, stating it needs to know more about the user
# okay let's give X-D
# this identifies all changes-commits by username@hostname
# (usefull in teams to track: what user from what machine did those changes)
git config --global user.name "user"
git config --global user.email "user@$(hostname)"

# create a commit
git commit -m "the_very_beginning" -a
# connect the server git repository to client
# if 1234 is git-server-ssh-port (change)
git remote add origin ssh://user@server.com:1234/home/user/projects/test.git

# or if "target" is on the same system as development (everything local) add "local" origin like
git remote add origin /home/user/projects/test.git

# optional: error-correction: if the origin needs to be modified
# it first needs to be deleted, then re-added
git remote remove origin

# on client: push/sync local changes to server
git push origin master

# then try again
git remote add origin ssh://user@server.com:1234/home/user/projects/test.git

git push origin master

### HURRAY! :) ONE JUST CREATED A NEW GIT REPO!
# AND SYNCED CHANGES TO IT! :)
# CONGRATULATIONS ALL INVOLVED!


# now let's see if one can pull the changes
# (maybe on a different 3rd (virtual)machine?)
# (anything but localhost works)

# on client: create new test dir
mkdir -p ~/projects/test.clone.git
cd ~/projects/test.clone.git

# pull repo to local folder (folder test will be automatically created)
git clone ssh://user@server.com:1234/home/user/projects/test.git

Cloning into 'test'...
remote: Enumerating objects: 3, done.
remote: Counting objects: 100% (3/3), done.
remote: Compressing objects: 100% (2/2), done.
remote: Total 3 (delta 0), reused 0 (delta 0)
Receiving objects: 100% (3/3), done.

# let's see what one got
ll

total 28K
drwxr-xr-x   3 user user 4.0K Apr 25 13:29 .
drwxr-xr-x 357 user user  20K Apr 25 13:27 ..
drwxr-xr-x   3 user user 4.0K Apr 25 13:29 test

ll test
total 16K
drwxr-xr-x 3 user user 4.0K 2020-04-25 .
drwxr-xr-x 3 user user 4.0K 2020-04-25 ..
drwxr-xr-x 8 user user 4.0K 2020-04-25 .git
-rw-r--r-- 1 user user   52 2020-04-25 README.txt

cat ./test/README.txt
this is a test git repository created on 2020-04-25

### HURRAY! one just successfully pulled the project from server to client :)

# view status of this git repo (what branch)
git log

# to change the location on server or the server
git remote remove origin
git remote add origin ssh://user@newserver.com:12345/new/location

how to clone the repo?

“cloning” aka downloading the files inside the repo to the local machine works like this:

git clone ssh://user@server.com:1234/home/user/projects/test.git

how to update local?

before changing files!

always pull the latest changes from git-srv! 🙂

this can easily be forgotten…

clone yesterday… work-make-changes today… argh… local files not up to date and might need merging

in order to refresh the local files with the latest changes from git-srv cd into the project folder then:

git pull origin master

PS: git pull origin master does NOT delete newly created local files

let’s script it:

its always more convenient, to just run /scripts/do_stuff.sh ProjectName

than to actually go through each and every necessary command

NOTE!

instead of ~/projects one has (unwisely) chosen /projects (subfolder of root) (please modify X-D)

server:

create new project script

vim /scripts/git.server.add_project.sh

#!/bin/bash
echo "===== create new dir for project ====="
mkdir -p ~/projects/$1

echo "===== change into that dir ====="
cd ~/projects/$1

echo "===== git init --bare ====="
git init --bare

client:

note: my project folder is /projects not /home/user/projects (which might be an better idea depending on how one runs backups)

git.client.add_project.sh

the idea is to create an empty local folder and git repo (which this script generates)

add folders

commit them (commit itself does not upload the files, git push does)

need to be changed: ip and username and port of ssh-git-server

cat /scripts/git.client.add_project.sh

#!/bin/bash
echo "=== create repository directory (where to store project on client) ==="
mkdir -p /projects/$1
cd /projects/$1

echo "=== remove exisint maybe old .git dir if present ==="
rm -rf .git

echo "=== initialize the client folder as git repository ==="
git init

echo "=== make git realize there are new files ==="
git add .

echo -e "this repository was created $(date '+%Y-%m-%d-%H:%M:%S')\n\r and this file is the first/initial commit" > README.txt

echo "=== create a commit ==="
git commit -m "first-initial commit, added README.txt" -a

echo "=== connect the server git repository to client ==="
# 2223 is the port ssh is running on
git remote add origin ssh://$USER@123.ip.of.server:port1234/home/$USER/projects/$1

echo "=== push/sync local changes to server ==="
git push origin master

download = clone a repository server2local:

to download a repo from server2local is called to clone

if ther user pulls a repo it is not cloned (no files visible)

cat /scripts/git.download.project.sh

#!/bin/bash
echo "=== create repository directory (where to store project on client) ==="
mkdir -p /projects/$1
cd /projects/$1

echo "=== remove exisint maybe old .git dir if present ==="
rm -rf .git

echo "=== initialize the client folder as git repository ==="
git init

echo "=== make git realize there are new files ==="
git add .

echo -e "this repository was created $(date '+%Y-%m-%d-%H:%M:%S')\n\r and this file is the first/initial commit" > README.txt

echo "=== connect the server git repository to client ==="
# port1234 is the port ssh is running on
git remote add origin ssh://$USER@123.ip.of.server:port1234/home/$USER/projects/$1

echo "=== push/sync local changes to server ==="
git clone ssh://$USER@123.ip.of.server:port1234/home/$USER/projects/$1

upload/commit changes to server script:

this script is meant that when changes were made to the empty local repo

to upload = commit them

before changes can be uploaded “commited”

it is required to identify the user

in order to track, what user made what changes

# set per-project identity info
git config --global user.email "user@domain.com"
git config --global user.name "user"

cat /scripts/git.client.commit.sh

#!/bin/bash
echo "=== sync changes from client to master git repo@server ==="
echo "= project: $1 ="

echo "= changing directory ="
cd $1

echo "= adding files ="
git add .

echo "= committing changes ="
git commit -m "$2"

echo "= uploading changes ="
git push origin master

# usage
/scripts/git.client.commit.sh local_project_folder "commit message"

git digital workflow in a nice chart:

creditz: https://bytes.inso.cc/2009/04/22/git-presentation-at-purple-scout/git-repository-diagram/

tips + hints:

if the initial files files for the repo are on a a server and the local empty folder shall become a git repo, the user might encouter this error

git.error: “There is no tracking information for the current branch. Please specify which branch you want to merge with. See git-pull(1) for details.”

simply instead of

git pull

git pull origin master

git.error: “error: src refspec main does not match any”

this means there is no such branch in the server’s repository

if the user tried: git push origin main

try it with: git push origin master

(unfortunately, there does not seem to be a default branch name definition? with github using “main”)

common (everyday) git problems and pitfalls:

during git add . “fatal: detected dubious ownership in repository at”
- “happens when a repo was cloned by a different user than your current user” (src)
- but git also delivers a possible workaround in the error message:
- To add an exception for this directory, call:
```
git config --global --add safe.directory ~/projects/name-of-git-project-folder
```

PS:

man git: git.man.txt “git – the stupid content tracker”