AWS Security Blog

How to Implement Federated API and CLI Access Using SAML 2.0 and AD FS

Feb 21 2023: This post is out of date. AWS now recommends using IAM Identity Center for federated identities accessing AWS by the CLI. Please see this post for more info: AWS CLI v2 Preview Now Supports AWS Single Sign-On

Important note from July 18, 2019: The original version of this blog uses Python2.x scripts. We now have a Python3.x script that you can download here: Python3.x script

Note from May 24, 2019: The features and services described in this post have changed since the post was published and the procedures described might be out of date and no longer accurate. Specifically, the procedures have not been tested with Python 3.x. If we update this post or create a replacement, we’ll add a notification about it here.


Note 1: On August 12, 2015, I published a follow-up to this post, which is called How to Implement a General Solution for Federated API/CLI Access Using SAML 2.0. Be sure to see that post if you want to implement a general federation solution (not specific to AD FS).

Note 2: This post focuses on NTLM authentication, the default authentication mechanism for AD FS 2.0. If you are using AD FS 3.0—which uses form-based authentication by default—see How to Implement a General Solution for Federated API/CLI Access Using SAML 2.0.


AWS supports identity federation using SAML (Security Assertion Markup Language) 2.0. Using SAML, you can configure your AWS accounts to integrate with your identity provider (IdP). Once configured, your federated users are authenticated and authorized by your organization’s IdP, and then can use single sign-on (SSO) to sign in to the AWS Management Console. This not only obviates the need for your users to remember yet another user name and password, but it also streamlines identity management for your administrators. This is great if your federated users want to access the AWS Management Console, but what if they want to use the AWS CLI or programmatically call AWS APIs?

In this blog post, I will show you how you can implement federated API and CLI access for your users. The examples provided use the AWS Python SDK and some additional client-side integration code. If you have federated users that require this type of access, implementing this solution should earn you more than one high five on your next trip to the water cooler.

Let’s start by taking a quick look at where we’ll end up.

janedoe@Ubuntu64:/tmp$ ./samlapi.py 
Username: ADjanedoe
Password: **************** 

Please choose the role you would like to assume:
[ 0 ]:  arn:aws:iam::012345678987:role/ADFS-Administrators
[ 1 ]:  arn:aws:iam::012345678987:role/ADFS-Operators
Selection:  1


---------------------------------------------------------------
Your new access key pair has been stored in the aws configuration 
file /home/janedoe/.aws/credentials under the saml profile.
Note that it will expire at 2015-05-26T17:16:20Z.
After this time you may safely rerun this script to refresh your 
access key pair.
To use this credential call the aws cli with the --profile option 
(e.g. aws --profile saml ec2 describe-instances).
---------------------------------------------------------------

Simple API example listing all s3 buckets:
[<Bucket: mybucket1>, <Bucket: mybucket2>, <Bucket: mybucket3>, 
<Bucket: mybucket4>, <Bucket: mybucket5>]

What are we seeing in the preceding output exactly?

  1. The utility prompts a federated user to enter her Active Directory credentials. These credentials are used to securely authenticate and authorize her against the configured IdP.
  2. The utility inspects the returned SAML assertion and determines the IAM roles she has been authorized to assume. After she selects her desired role, the utility uses the AWS Security Token Service (STS) to retrieve temporary security credentials.
  3. The utility automatically writes these credentials to her local AWS credentials file and she can begin issuing AWS API or CLI calls.

After taking this sample and customizing it for your organization, you’ll be able to leverage the power and automation of the AWS API and CLI interfaces by using your organization’s credentials, while maintaining the control provided by the AWS Identity and Access Management (IAM) service.

This post will focus on using Microsoft Active Directory Federation Services (AD FS), but if you are using a different provider, don’t despair because the basic pattern should work with other common IdPs.

Getting started

To follow along with this post, you must have:

  1. AD FS correctly integrated with your AWS account for console access using only your organizational credentials. See Enabling Federation to AWS using Windows Active Directory, ADFS, and SAML 2.0, if you need instructions on setting this up.
  2. A recent version (2.36+) of the AWS Python SDK installed on your local workstation.
  3. A minimal AWS credentials file (for example, ~/.aws/credentials) with the following contents adjusted to your preferred region and output format.
[default]
output = json
region = us-west-2
aws_access_key_id =
aws_secret_access_key =

Important note #1: An AWS access key pair is not configured in the above structure because the initial AWS STS call is authenticated by the SAML assertion returned by the trusted IdP. All subsequent API/CLI calls are authenticated by the key pair contained within the returned AWS STS token. For additional details, see Giving AWS Console Access to Federated Users Using SAML.

Important note #2: In the past, the various AWS SDK and CLI packages supported both config and credential files. These files have similar but slightly different syntaxes. Credential files are now standard and will be used in this post.  If you are still using a config file, you’ll need to convert to a credential file in order to avoid errors later in the process.

First, you need to install two of the modules that fall outside the core Python distribution, specifically beautifulsoup4 and requests-ntlm. There are several ways to do this, but the pip utility, which is included in Python 2.7.9+, makes it a very easy task. You simply need to run the following two commands:

pip install beautifulsoup4

and then

pip install requests-ntlm

You should get an output that looks similar to the following screenshot.

Image of the output

The AWS federation process utilizes what is known as an IdP-initiated login. In the last preparatory step, you’ll need to determine the specific URL that is being used to initiate this login. Start with your basic IdP-initiated login URL (the one you use for federated access to your various SAML Relying Parties, including the AWS Management Console). In this example, I’m using AD FS 2.0, where the URL takes the form of https://<fqdn>/adfs/ls/IdpInitiatedSignOn.aspx. If I place this URL in a browser’s address field, I’m presented with a site selection page, as shown in the following image.

Image of site selection page

In order to construct the URL you need, concatenate the original IdP-initiated login URL with the query string ?loginToRp=urn:amazon:webservices. The resulting URL should take the form of https://<fqdn>/adfs/ls/IdpInitiatedSignOn.aspx?loginToRp=urn:amazon:webservices. Save this URL because it will be referred to as the idpentryurl variable below. If you place this concatenated URL in a browser’s address field you should still reach the AD FS IdP-initiated login process. However, you bypass the site selection page and proceed straight to the AWS Management Console after you authenticate, which is the URL you need for this exercise. If you are interested, this query string is documented on Microsoft’s website.

Important note: Be sure to pay close attention to the capitalization of the hostname in the URL. AD FS uses this as part of the authentication process, and any mismatch in capitalization will cause the authentication to fail.

Reviewing the code

Now that you have gone through the steps in the “Getting Started” section above, you are ready to start assembling the integration code that enables federated API and CLI access. I’ll explain what’s going on one chunk at a time so that you can follow along, and I have included a link to the full code example at the end of this post. First, let’s set up some basic imports and variables.

#!/usr/bin/python 
 
import sys 
import boto.sts 
import boto.s3 
import requests 
import getpass 
import ConfigParser 
import base64 
import xml.etree.ElementTree as ET 
from bs4 import BeautifulSoup 
from os.path import expanduser 
from urlparse import urlparse, urlunparse 
from requests_ntlm import HttpNtlmAuth
 
##########################################################################
# Variables 
 
# region: The default AWS region that this script will connect 
# to for all API calls 
region = 'us-west-2' 
 
# output format: The AWS CLI output format that will be configured in the 
# saml profile (affects subsequent CLI calls) 
outputformat = 'json'
 
# awsconfigfile: The file where this script will store the temp 
# credentials under the saml profile 
awsconfigfile = '/.aws/credentials'
 
# SSL certificate verification: Whether or not strict certificate 
# verification is done, False should only be used for dev/test 
sslverification = True 
 
# idpentryurl: The initial URL that starts the authentication process. 
idpentryurl = 'https://<fqdn>/adfs/ls/IdpInitiatedSignOn.aspx?loginToRp=urn:amazon:webservices' 
 
##########################################################################

Adjust the variables in the previous code to suit your specific region (e.g., us-west-2, us-east-1, etc.) and format preferences (i.e., json, text, or table), and then plug in the value noted from the previous section in this post for the idpentryurl. If you’re using a self-signed certificate in a trusted testing environment, you’ll also need to set sslverification to False, in order to avoid certificate verification failures.

With the basic environment set up, prompt the user and retrieve the user’s credentials from standard input. I’ll talk later in this post about how we can support other credential forms.

# Get the federated credentials from the user
print "Username:",
username = raw_input()
password = getpass.getpass()
print ''

The comma following the first print ”Username”, statement might look like a bug, but it is actually a little Python trick to prevent the print statement from appending a newline character. You’ll also note the use of the getpass() method, which prevents the password from being echoed out in the output.

Next, we use the Python requests module to assemble the authentication information and formulate the https request to the IdP as shown in the following snippet. Assuming the authentication is successful, the response returned by AD FS will contain the SAML assertion

# Initiate session handler 
session = requests.Session() 
 
# Programatically get the SAML assertion 
# Set up the NTLM authentication handler by using the provided credential 
session.auth = HttpNtlmAuth(username, password, session) 
 
# Opens the initial AD FS URL and follows all of the HTTP302 redirects 
response = session.get(idpentryurl, verify=sslverification) 
 
# Debug the response if needed 
#print (response.text)

At this point we have the IdP response safely in hand, so let’s overwrite and dereference the variables we used to store the user’s username and password because they are no longer needed. Python, as a higher-level language, doesn’t provide the direct memory manipulation to do this like one might in C, but this does a good job of ensuring the credentials don’t linger around.

# Overwrite and delete the credential variables, just for safety
username = '##############################################'
password = '##############################################'
del username
del password

As I noted previously, the SAML assertion that we need can be found in the response from the IdP. The following code uses the BeautifulSoup module to extract it.

# Decode the response and extract the SAML assertion 
soup = BeautifulSoup(response.text.decode('utf8')) 
assertion = '' 
 
# Look for the SAMLResponse attribute of the input tag (determined by 
# analyzing the debug print lines above) 
for inputtag in soup.find_all('input'): 
    if(inputtag.get('name') == 'SAMLResponse'): 
        #print(inputtag.get('value')) 
        assertion = inputtag.get('value')

This snippet simply iterates through all of the HTML input tags until it finds the one named SAMLResponse. The Base64-encoded assertion we’re after is contained in the value attribute of this tag. Before moving on, we need to go one layer deeper into the SAML authentication response itself and extract the list of the user’s authorized roles. This list is stored in the SAML attribute named https://aws.amazon.com/SAML/Attributes/Role, as shown in the following code.

# Parse the returned assertion and extract the authorized roles 
awsroles = [] 
root = ET.fromstring(base64.b64decode(assertion))
 
for saml2attribute in root.iter('{urn:oasis:names:tc:SAML:2.0:assertion}Attribute'): 
    if (saml2attribute.get('Name') == 'https://aws.amazon.com/SAML/Attributes/Role'): 
        for saml2attributevalue in saml2attribute.iter('{urn:oasis:names:tc:SAML:2.0:assertion}AttributeValue'):
            awsroles.append(saml2attributevalue.text)
 
 
 
# Note the format of the attribute value should be role_arn,principal_arn 
# but lots of blogs list it as principal_arn,role_arn so let's reverse 
# them if needed 
for awsrole in awsroles: 
    chunks = awsrole.split(',') 
    if'saml-provider' in chunks[0]:
        newawsrole = chunks[1] + ',' + chunks[0] 
        index = awsroles.index(awsrole) 
        awsroles.insert(index, newawsrole) 
        awsroles.remove(awsrole)

In the first loop above, I’m using the Python xml.etree.ElementTree module to parse the SAML assertion. Though assertions aren’t particularly pretty to look at, the specific namespaces and names make finding what you’re looking for relatively straightforward.

In the second loop, the code is just doing a bit of proactive error proofing to make sure the Amazon Resource Names (ARNs) contained within the roles are in the right order.

Now we know the roles the user is authorized to assume. Next, the code asks the user which role they want to assume.

# If I have more than one role, ask the user which one they want, 
# otherwise just proceed 
print "" 
if len(awsroles) > 1: 
    i = 0 
    print "Please choose the role you would like to assume:" 
    for awsrole in awsroles: 
        print '[', i, ']: ', awsrole.split(',')[0] 
        i += 1 

    print "Selection: ", 
    selectedroleindex = raw_input() 
 
    # Basic sanity check of input 
    if int(selectedroleindex) > (len(awsroles) - 1): 
        print 'You selected an invalid role index, please try again' 
        sys.exit(0) 
 
    role_arn = awsroles[int(selectedroleindex)].split(',')[0] 
    principal_arn = awsroles[int(selectedroleindex)].split(',')[1]
 
else: 
    role_arn = awsroles[0].split(',')[0] 
    principal_arn = awsroles[0].split(',')[1]

If the user only has one role, it is automatically used. Otherwise, the user will be prompted to select which role they want to use from a simple enumerated list.

Things now start to get quite a bit easier. Because we’ve already retrieved the SAML assertion and extracted the various ARNs, we just need to call the standard AWS STS service to request AWS temporary security credentials. This is really the crux of the entire utility. We’ve undertaken all of the effort up to this point to prepare the parameters for the following simple AssumeRolewithSAML API call.

# Use the assertion to get an AWS STS token using Assume Role with SAML
conn = boto.sts.connect_to_region(region)
token = conn.assume_role_with_saml(role_arn, principal_arn, assertion)

Important note: The SAML assertion, which comes from a configured and trusted IdP, serves as the authentication credential for this API call.

Now that we have our credentials from AWS STS, the next step is to insert these credentials into a specific profile within the AWS credentials file as shown in the following code snippet. By doing so, you’ll be able to use the temporary security credentials for any number of API or CLI calls until the credentials expire.

# Write the AWS STS token into the AWS credential file
home = expanduser("~")
filename = home + awsconfigfile
 
# Read in the existing config file
config = ConfigParser.RawConfigParser()
config.read(filename)
 
# Put the credentials into a specific profile instead of clobbering
# the default credentials
if not config.has_section('saml'):
    config.add_section('saml')
 
config.set('saml', 'output', outputformat)
config.set('saml', 'region', region)
config.set('saml', 'aws_access_key_id', token.credentials.access_key)
config.set('saml', 'aws_secret_access_key', token.credentials.secret_key)
config.set('saml', 'aws_session_token', token.credentials.session_token)
 
# Write the updated config file
with open(filename, 'w+') as configfile:
    config.write(configfile)

In this code snippet, I’m simply using a combination of variables to locate the AWS credentials file, reading in the file, adding or updating a profile called ‘saml’, and writing the file back out.

Finally, the code gives the federated user a basic idea of what has just happened.

# Give the user some basic info as to what has just happened
 
print 'nn----------------------------------------------------------------'
 
print 'Your new access key pair has been stored in the AWS configuration file {0} under the saml profile.'.format(filename)
 
print 'Note that it will expire at {0}.'.format(token.credentials.expiration)
 
print 'After this time you may safely rerun this script to refresh your access key pair.'
 
print 'To use this credential call the AWS CLI with the --profile option (e.g. aws --profile saml ec2 describe-instances).'
 
print '----------------------------------------------------------------nn'

If your automation is CLI based, you can skip ahead to the next section, “Putting it all together.” Alternatively, if your automation is API based, you can simply use the temporary security credentials when making your connections.

# Use the AWS STS token to list all of the S3 buckets
s3conn = boto.s3.connect_to_region(region,
                     aws_access_key_id=token.credentials.access_key,
                     aws_secret_access_key=token.credentials.secret_key,
                     security_token=token.credentials.session_token)
 
buckets = s3conn.get_all_buckets()
 
print 'Simple API example listing all s3 buckets:'
print(buckets)

In practice, it would be far better form to place these additional API calls in a separate script that utilizes the credential profile we set up above, but the point of the exercise is that the credentials received from AWS STS are being used to authenticate the API call.

Putting it all together

When you put all of this code together, you see what’s shown in the following output.

janedoe@Ubuntu64:/tmp$ ./samlapi.py
Username: ADjanedoe
Password: ****************
 
Please choose the role you would like to assume:
[ 0 ]:  arn:aws:iam::012345678987:role/ADFS-Administrators
[ 1 ]:  arn:aws:iam::012345678987:role/ADFS-Operators
Selection:  1
 
 
---------------------------------------------------------------Your new access key pair has been stored in the aws configuration file /home/janedoe/.aws/credentials under the saml profile.
Note that it will expire at 2015-05-26T17:16:20Z.
After this time you may safely rerun this script to refresh your access key pair.
To use this credential call the aws cli with the --profile option (e.g. aws --profile saml ec2 describe-instances).
---------------------------------------------------------------
 
Simple API example listing all s3 buckets:
[<Bucket: mybucket1>, <Bucket: mybucket2>, <Bucket: mybucket3>, <Bucket: mybucket4>, <Bucket: mybucket5>]

 

What are we seeing in the preceding output exactly?

  1. The utility prompts a federated user to enter her Active Directory credentials. These credentials are used to securely authenticate and authorize her against the configured IdP.
  2. The utility inspects the returned SAML assertion and determines the IAM roles she has been authorized to assume.  After she selects her desire role, the utility uses the AWS Security Token Service (STS) to retrieve temporary security credentials.
  3. The utility automatically writes these credentials to her local AWS credentials file and she can begin issuing AWS API or CLI calls.

You can then issue subsequent API or CLI calls by simply referencing the saml profile, as shown in the following output.

janedoe@Ubuntu64:/tmp$ aws --profile saml --output text s3api list-buckets
BUCKETS   2014-09-09T17:11:48.000Z     mybucket1
BUCKETS   2014-11-03T18:27:35.000Z     mybucket2
BUCKETS   2014-08-13T02:45:28.000Z     mybucket3
BUCKETS   2015-01-14T17:53:47.000Z     mybucket4
BUCKETS   2015-05-19T19:23:25.000Z     mybucket5
OWNER     myowner   128abcedf111aa33cc44ddee5512345abcd6fff4eed

In the end, the browser oriented nature of SAML requires a bit of code and bit of engineering, but it accomplishes the goal: enabling federation-based authentication for API and CLI calls!

Using Other Identity Providers or Authentication Mechanisms

If you are using an IdP other than AD FS, don’t despair that your solution was not included in this blog post. The reason I chose the integration technique of using a Python module to emulate browser behavior is that it should be extensible to just about any IdP solution.

Similarly, if you are using other authentication mechanisms such as Kerberos, X.509 certificates, or Multi-Factor Authentication, there are additional handlers within the Python requests module. These additional handlers can be used to enhance the example above after you have the basics working.

If you want to get started quickly, see the full code example for Python2.x and the full code example for Python3.x. I hope you enjoyed this post, and I look forward to your feedback.

– Quint

Want more AWS Security how-to content, news, and feature announcements? Follow us on Twitter.