[ci skip]

-s flag of su doesn't work correctly on some systems, loging in the user
and not running the requested command. By moving the bash shell init
inside the su command we avoid the issue

Closes https://github.com/gitlabhq/gitlabhq/issues/9869

[ci skip]

It turns out that if we do not the declaration from
"location /" wins.

Limiting, if any, should happen in gitlab-workhorse.

- Offloads uploading to GitLab Workhorse
- Use /authorize request for fast uploading
- Added backup recipes for artifacts
- Support download acceleration using X-Sendfile

Users are allowed to supply namespace%2Fproject instead of a numeric ID

This change relies on changes in gitlab_git and gitlab-git-http-server.

Before this change NGINX would convert a chunked HTTP POST (e.g.
git push) into a HTTP 1.0 single large POST. This creates an
unnecessary delay, and it creates unnecessary memory pressure on
gitlab-git-http-server.

For the response ('proxy_buffering') I am less sure that NGINX 's
buffering behavior is harmful, but it still makes more sense to me
not to interfere with gitlab-git-http-server (and the Golang net/http
server).

https://gitlab.com/gitlab-org/gitlab-git-http-server

This change introduces the GITLAB_GRACK_AUTH_ONLY environment
variable. When set, Grack requests to GitLab will only respond with
the user's GL_ID (if the request is OK) or an error. This allows
gitlab-git-http-server to use the main GitLab application as an
authentication and authorization backend.

If we like how this works we should drop the GITLAB_GRACK_AUTH_ONLY
variable at some point in the future.

- Use whoami instead of relying on the existence of $USER enviroment variable which is not always present

Revert "Merge branch 'go-get-workaround-nginx' of https://github.com/mattes/gitlabhq into mattes-go-get-workaround-nginx"

This reverts commit 51349ca3, reversing
changes made to b180476b.