Login to Gitlab with username and password using cURL

Question

For testing a command line tool against a Gitlab instance in Docker, I would like to login to Gitlab using username and password and grab the created session to authenticate my API requests.

Therefore I do the following:

1. curl the user login page with curl -i http://localhost:8080/users/sign_in -s
2. get me the _gitlab_session from the headers
3. get me the authenticity_token from the login form
4. send a second curl request with

curl 'http://localhost:8080/users/sign_in' \
  -H "_gitlab_session=${cookie}" \
  -H 'Origin: http://localhost:8080' \
  -H 'Content-Type: application/x-www-form-urlencoded' \
  -H 'Accept: text/html,application/xhtml+xml,application/xml;q=0.9,image/webp,image/apng,*/*;q=0.8' \
  -H 'Referer: http://localhost:8080/users/sign_in' \
  -H 'Connection: keep-alive' \
  --data-urlencode "authenticity_token=${token}" \
  --data-urlencode "user[login]=root" \
  --data-urlencode "user[password]=12341234" \
  --data-urlencode "user[remember_me]=0"

But instead of a valid user login, I get a

422 - The change you requested was rejected.

And in the logfiles, I see

==> gitlab-rails/production.log <==
Started POST "/users/sign_in" for 172.17.0.1 at 2017-12-23 00:22:16 +0000
Processing by SessionsController#create as HTML
Parameters: {"authenticity_token"=>"[FILTERED]", "user"=>{"login"=>"root", "password"=>"[FILTERED]", "remember_me"=>"0"}}
Can't verify CSRF token authenticity
Completed 422 Unprocessable Entity in 125ms (ActiveRecord: 7.8ms)

==> gitlab-rails/production_json.log <==
{"method":"POST","path":"/users/sign_in","format":"html","controller":"SessionsController",  
"action":"create","status":422,"error":" 
ActionController::InvalidAuthenticityToken:ActionController::InvalidAuthenticityToken",
"duration":126.29,"view":0.0,"db":7.78,"time":"2017-12-23T00:22:16.039Z",
"params":{"authenticity_token":"[FILTERED]","user":{"login":"root","password":"
[FILTERED]","remember_me":"0"}},"remote_ip":"172.17.0.1",
"user_id":1,"username":"root"}

==> gitlab-rails/production.log <==

ActionController::InvalidAuthenticityToken (ActionController::InvalidAuthenticityToken):
  lib/gitlab/middleware/multipart.rb:93:in `call'
  lib/gitlab/request_profiler/middleware.rb:14:in `call'
  lib/gitlab/middleware/go.rb:18:in `call'
  lib/gitlab/etag_caching/middleware.rb:11:in `call'
  lib/gitlab/middleware/read_only.rb:31:in `call'
  lib/gitlab/request_context.rb:18:in `call'
  lib/gitlab/metrics/requests_rack_middleware.rb:27:in `call'

I assume that I forgot to pass a necessary ingredient in the second request - but I can't figure out, which one.

Answer 1

With some helpful tips by other answers and comments, I finally came up with this solution:

gitlab_host="http://localhost:8080"
gitlab_user="root"
gitlab_password="12341234"

# curl for the login page to get a session cookie and the sources with the auth tokens
body_header=$(curl -c cookies.txt -i "${gitlab_host}/users/sign_in" -s)

# grep the auth token for the user login for
#   not sure whether another token on the page will work, too - there are 3 of them
csrf_token=$(echo $body_header | perl -ne 'print "$1\n" if /new_user.*?authenticity_token"[[:blank:]]value="(.+?)"/' | sed -n 1p)

# send login credentials with curl, using cookies and token from previous request
curl -b cookies.txt -c cookies.txt -i "${gitlab_host}/users/sign_in" \
    --data "user[login]=${gitlab_user}&user[password]=${gitlab_password}" \
    --data-urlencode "authenticity_token=${csrf_token}"

# send curl GET request to personal access token page to get auth token
body_header=$(curl -H 'user-agent: curl' -b cookies.txt -i "${gitlab_host}/profile/personal_access_tokens" -s)
csrf_token=$(echo $body_header | perl -ne 'print "$1\n" if /authenticity_token"[[:blank:]]value="(.+?)"/' | sed -n 1p)

# curl POST request to send the "generate personal access token form"
# the response will be a redirect, so we have to follow using `-L`
body_header=$(curl -L -b cookies.txt "${gitlab_host}/profile/personal_access_tokens" \
    --data-urlencode "authenticity_token=${csrf_token}" \
    --data 'personal_access_token[name]=golab-generated&personal_access_token[expires_at]=&personal_access_token[scopes][]=api')

# Scrape the personal access token from the response HTML
personal_access_token=$(echo $body_header | perl -ne 'print "$1\n" if /created-personal-access-token"[[:blank:]]value="(.+?)"/' | sed -n 1p)

According to the GitLab API documentation you can now use the session cookie to authenticate API requests:

curl --header "Private-Token: ${personal_access_token}" https://gitlab.example.com/api/v4/projects

Some hints:

• I first got confused with the curl -c file (which reads cookies from the headers and writes them to the file) vs. the curl -b file (which uses the cookies in the file and sends them with the request)
• Unfortunately I did not find a regex that worked with sed so I have to use perl here.
• Using the developer console in Chrome and copying a POST request as curl command was very helpful: https://www.alexkras.com/copy-any-api-call-as-curl-request-with-chrome-developer-tools/

Answer 2

what is in ${cookie} exactly? in my tests, there were 4 cookies from the login page, 3 of which looks like tokens, it wouldn't surprise me if at least 1 more cookie other than the _gitlab_session cookie was required for logging in successfully. but instead of bothering to find the correct combination of required cookies, i think it would be better to just let curl handle cookies automatically, eg --cookie-jar and --cookie (then all of the cookies should be correct) - further more, there is 3 different authenticity_token's on the login page, are you sure you fetch the correct one? maybe your problem is that you're fetching the wrong token. differentiating between the 3 tokens is so difficult that i recommend resorting to a scripting language instead.

here's a tested working example in PHP using hhb_curl (just replace the username and password on line 10 & 11):

<?php
declare(strict_types = 1);
require_once ('hhb_.inc.php');
$hc = new hhb_curl ( '', true );
$html = $hc->exec ( 'https://gitlab.com/users/sign_in' )->getStdOut ();
$domd = @DOMDocument::loadHTML ( $html );
$inputs = getDOMDocumentFormInputs ( $domd, true, false ) ['new_user'];
// var_dump ( $inputs );
$inputs ['user[login]'] = '???';
$inputs ['user[password]'] = '???';
$html = $hc->setopt_array ( array (
        CURLOPT_POST => 1,
        CURLOPT_POSTFIELDS => http_build_query ( $inputs ) 
) )->exec ()->getStdOut ();
echo $html;
$domd = @DOMDocument::loadHTML ( $html );
$xp = new DOMXPath ( $domd );
foreach ( $xp->query ( '//div[contains(@class,"flash-alert")]' ) as $loginError ) {
    var_dump ( "login error: ", trim($loginError->textContent) );
}

function getDOMDocumentFormInputs(\DOMDocument $domd, bool $getOnlyFirstMatches = false, bool $getElements = true): array {
    // :DOMNodeList?
    if (! $getOnlyFirstMatches && ! $getElements) {
        throw new \InvalidArgumentException ( '!$getElements is currently only implemented for $getOnlyFirstMatches (cus im lazy and nobody has written the code yet)' );
    }
    $forms = $domd->getElementsByTagName ( 'form' );
    $parsedForms = array ();
    $isDescendantOf = function (\DOMNode $decendant, \DOMNode $ele): bool {
        $parent = $decendant;
        while ( NULL !== ($parent = $parent->parentNode) ) {
            if ($parent === $ele) {
                return true;
            }
        }
        return false;
    };
    // i can't use array_merge on DOMNodeLists :(
    $merged = function () use (&$domd): array {
        $ret = array ();
        foreach ( $domd->getElementsByTagName ( "input" ) as $input ) {
            $ret [] = $input;
        }
        foreach ( $domd->getElementsByTagName ( "textarea" ) as $textarea ) {
            $ret [] = $textarea;
        }
        foreach ( $domd->getElementsByTagName ( "button" ) as $button ) {
            $ret [] = $button;
        }
        return $ret;
    };
    $merged = $merged ();
    foreach ( $forms as $form ) {
        $inputs = function () use (&$domd, &$form, &$isDescendantOf, &$merged): array {
            $ret = array ();
            foreach ( $merged as $input ) {
                // hhb_var_dump ( $input->getAttribute ( "name" ), $input->getAttribute ( "id" ) );
                if ($input->hasAttribute ( "disabled" )) {
                    // ignore disabled elements?
                    continue;
                }
                $name = $input->getAttribute ( "name" );
                if ($name === '') {
                    // echo "inputs with no name are ignored when submitted by mainstream browsers (presumably because of specs)... follow suite?", PHP_EOL;
                    continue;
                }
                if (! $isDescendantOf ( $input, $form ) && $form->getAttribute ( "id" ) !== '' && $input->getAttribute ( "form" ) !== $form->getAttribute ( "id" )) {
                    // echo "this input does not belong to this form.", PHP_EOL;
                    continue;
                }
                if (! array_key_exists ( $name, $ret )) {
                    $ret [$name] = array (
                            $input 
                    );
                } else {
                    $ret [$name] [] = $input;
                }
            }
            return $ret;
        };
        $inputs = $inputs (); // sorry about that, Eclipse gets unstable on IIFE syntax.
        $hasName = true;
        $name = $form->getAttribute ( "id" );
        if ($name === '') {
            $name = $form->getAttribute ( "name" );
            if ($name === '') {
                $hasName = false;
            }
        }
        if (! $hasName) {
            $parsedForms [] = array (
                    $inputs 
            );
        } else {
            if (! array_key_exists ( $name, $parsedForms )) {
                $parsedForms [$name] = array (
                        $inputs 
                );
            } else {
                $parsedForms [$name] [] = $tmp;
            }
        }
    }
    unset ( $form, $tmp, $hasName, $name, $i, $input );
    if ($getOnlyFirstMatches) {
        foreach ( $parsedForms as $key => $val ) {
            $parsedForms [$key] = $val [0];
        }
        unset ( $key, $val );
        foreach ( $parsedForms as $key1 => $val1 ) {
            foreach ( $val1 as $key2 => $val2 ) {
                $parsedForms [$key1] [$key2] = $val2 [0];
            }
        }
    }
    if ($getElements) {
        return $parsedForms;
    }
    $ret = array ();
    foreach ( $parsedForms as $formName => $arr ) {
        $ret [$formName] = array ();
        foreach ( $arr as $ele ) {
            $ret [$formName] [$ele->getAttribute ( "name" )] = $ele->getAttribute ( "value" );
        }
    }
    return $ret;
}

• it currently outputs

  string(13) "login error: " string(26) "Invalid Login or password."

meaning that ???/??? is not a valid username & password combination.

Answer 3

Create access_token or private_token in gitlab and you can use it in URL.

bash <(curl -s full-path-to-raw-url?private_token=your_private_token) param1 param2

Answer 4

I got a stable version working using Python, requests and beautifulsoup4 :

from urllib.parse import urljoin

import requests
import bs4

base_uri = 'http://gitlab.example.com/'
username = 'me'
password = 'let_me_in'

# Use a session so cookies are retained between requests.
session = requests.session()

# Load the login page to get a CSRF token.
response = session.get(urljoin(base_uri, 'users/sign_in'))
response.raise_for_status()

# Extract the CSRF token from the login page.
soup = bs4.BeautifulSoup(response.text, 'html.parser')
csrf_param = soup.find('meta', dict(name='csrf-param'))['content']
csrf_token = soup.find('meta', dict(name='csrf-token'))['content']

# Login.
request_data = {
    'username': username,
    'password': password,
    csrf_param: csrf_token}
response = session.post(response.url, data=request_data)
response.raise_for_status()

# Get a list of project.
response = session.get(urljoin(base_uri, 'api/v4/projects'))
print(response.json())

Answer 5

thanks to @Michael Lihs for his work. I am using gitlab V13 and I had to modify his solution to make it work for me. Here is what works for me :

gitlab_host="http://localhost:8080"
gitlab_user="root"
gitlab_password="12341234"
# step 1 create session and get authenticity_token
rm cookies.txt
body_header=$(curl -c cookies.txt -k -i "${gitlab_host}/users/sign_in" -s)
csrf_token=$(echo $body_header | perl -ne 'print "$1\n" if /new_user.*?authenticity_token"[[:blank:]]value="(.+?)"/' | sed -n 1p)

# step 2 identification
curl -b cookies.txt -c cookies.txt -L -i -k "${gitlab_host}/users/auth/ldapmain/callback" \
    -d username=$gitlab_user -d password="$gitlab_password" \
    --data-urlencode "authenticity_token=${csrf_token}"

# step 3 download files
rm /tmp/x3.pdf && curl -k "${gitlab_host}/my_group/my_project/-/raw/my_branch/my_fullpath_to_file" -b cookies.txt  --compressed > /tmp/x3.pdf