# How to assume an ECS task role in AWS, the official and the fake way

on
feb 27, 2021
in

When you work with AWS ECS and you apply the least-privileged security principle, you create an IAM role for each task. While developing the software, you may want to test the role too. However, this is not easily done. In this blog we will show you how to can be done in the official way, and in a slightly more alternative way with the iam-sudo utility for development purposes.
In our Cloud-Native setup principles, we do not allow anybody to make manual changes to the AWS environment. This
means that developers have view-only access even to the development environment. All the changes have to
be done through AWS CloudFormation.
Sometimes it is necessary that a developer gets to debug their software from their local laptop while
talking to the development environment. This requires them to be able to assume the role of the ECS task.

## What is the problem

Typically when you create an AWS ECS Task role, you grant the ecs-tasks service the permission to assume
the role. This means that only AWS ECS can assume the role, and nobody else. Let me illustrate this
through the following steps:
* create the role
* assume the role

### create the role

To create a role for an AWS ECS task, you grant the service ecs-tasks.amazonaws.com
the permission to so as follows:

aws iam create-role  \
--assume-role-policy-document '{
"Statement":  {
"Effect": "Allow",
"Principal": { "Service": [ "ecs-tasks.amazonaws.com" ] },
"Action": [ "sts:AssumeRole" ]
}]}'


Next you can add an inline policy to the role, for instance:

aws iam put-role-policy \
--policy-name 'describe-parameters' \
--policy-document '{
"Statement": [{
"Effect": "Allow",
"Action": [ "ssm:DescribeParameters" ],
"Resource": "*"
}]}'


### assume the role

To get the credentials to assume this role, you type:

ROLE_ARN=$(aws iam get-role \ --role-name iam-sudo-demo-my-task \ --query Role.Arn \ --output text) aws sts assume-role \ --role-arn$ROLE_ARN \
--role-session-name no-can-do


Unfortunately, this fails with the following error:

An error occurred (AccessDenied) when calling the AssumeRole operation:
User: arn:aws:iam::**********15:user/mvanholsteijn is not authorized
to perform: sts:AssumeRole on resource: arn:aws:iam::**********15:role/iam-sudo-demo-my-task


How come? I am root user of the AWS account! This is caused the absence of
the sts:AssumeRole permission for the AWS account itself.

## Possible solutions

There are at least two possible solutions to the problem:

1. extend the assume role document with the AWS account
2. fake the role using session policies

### extending the assume role to the account

One possible solution is to grant the AWS account the permission to assume the
role too, as follows:

ACCOUNT_ID=$(aws sts get-caller-identity \ --query Account --output text) aws iam create-role \ --role-name iam-sudo-demo-my-task-allow-root-assume-role \ --assume-role-policy-document "$(
jq -n . --arg account_id $ACCOUNT_ID '{ "Statement": [{ "Effect": "Allow", "Principal": { "Service": [ "ecs-tasks.amazonaws.com" ] }, "Action": [ "sts:AssumeRole" ] },{ "Effect": "Allow", "Principal": { "AWS": [$account_id ] },
"Action": [ "sts:AssumeRole" ]
}]}'
)"


The assume role will now work. It does require that developers are granted
permission to assume this role too. As we have many ECS task roles with CloudFormation
generated in separate cloudformation templates, this is quite hard. The easiest way to do this,
is to to tag the roles as follows:

aws iam create-role --tags Key=env,Value=dev ...


And grant a conditional assume role permission to the developers for these roles:

    "Statement": [{
"Effect": "Allow",
"Action": "sts:AssumeRole",
"Resource": "*",
"Condition": {"StringLike": {"iam:ResourceTag/env": "dev"}}
}]


Now, this is a solution that works entirely with the IAM solution space. The
only thing against this, is that it is quite an invasive change which is only
needed in the development environment.

### fake the role using session policies

The previous solution is the official solution. But, it adds
complexity which in our case, and is only required in the development environment.
When you look at the assume-role api call
you will see that when you assume a role, you can specify:
* the role to assume
* an inline session policy to apply
* one or more managed policies to apply
This means that when you read an IAM Role, you can apply the inline role policies,
the attached managed policies and apply them on a role that you are allowed
to assume.
Let me show you how this works. First create a role
that has sufficient permissions to still have an effective policy:

aws iam create-role  \
--role-name iam-sudo-demo-power-user \
--assume-role-policy-document "$(\ jq -n \ --arg account_id$ACCOUNT_ID \
'{
"Statement": [{
"Effect": "Allow",
"Principal": { "AWS": [ $account_id ] }, "Action": [ "sts:AssumeRole" ] }]}' \ )" aws iam attach-role-policy \ --role-name iam-sudo-demo-power-user --policy-arn arn:aws:iam::aws:policy/PowerUserAccess  Next we need to retrieve all of the inline session policies of the role: for NAME in$(aws iam list-role-policies \
--query PolicyNames --output text) ; \
do
aws iam get-role-policy \
--policy-name $NAME done | \ jq '.PolicyDocument.Statement' | \ jq -s add > inline-policies.json  To retrieve all of the managed policies attached to the role, type:  aws iam list-attached-role-policies \ --role-name iam-sudo-demo-my-task \ --query 'join(,, AttachedPolicies[*].join(=, [arn, PolicyArn]))' \ --output text \ | sed -e 's/^./--policy-arns &/' \ > attached-policies.txt  Finally, we can simulate the role: aws sts assume-role \ --role-arn arn:aws:iam::${ACCOUNT_ID}:role/iam-sudo-demo-power-user \
--policy "$(<inline-policies.json)" \$(<attached-policies.txt)


This still looks quite elaborate. That is why we have implemented this in the iam-sudo utility.

### using the IAM sudo utility

The IAM sudo utility simplifies the effort of
simulating a role. To do what is shown above, just type:

pip3 install iam-sudo

iam-sudo simulate --local \
--base-role iam-sudo-demo-power-user \
aws sts get-caller-identity

INFO: credentials saved under AWS profile iam-sudo-demo-my-task.
{
"Account": "**********15",
}


It saves the credentials under the specified profile name and executes the command with them.

### privilege elevation using AWS Lambda

Although simulating the role policies is nice, we still need to assume a role that is powerful
enough to render an effective session policy. In the example above, we assumed the role
with power user privileges. Giving this privilege defeats the whole purpose of
the exercise.
To address this issue, we created an AWS Lambda which will only hand out credentials based
on a simulated IAM role. Which roles you are allowed to assume is governed by a policy.
To install this lambda, type:

aws cloudformation create-stack \
--stack-name iam-sudo \
--template-url  https://binxio-public-eu-central-1.s3.amazonaws.com/lambdas/iam-sudo-0.2.0.yaml \
--capabilities CAPABILITY_NAMED_IAM
aws cloudformation wait stack-create-complete --stack-name iam-sudo


If you have the permission to invoke the Lambda, you can simulate the role using the
credentials provided by the Lambda:

iam-sudo simulate --remote \
aws sts get-caller-identity
{
"Account": "444093529715",
}


The deployed lambda will only generate credentials for roles for the service ecs-tasks and of which the role
name starts with iam-sudo-demo. You can change the policy

## Conclusion

To allow developers to assume any role in the development environment, requires
each of the roles to include the assume role permission for the AWS account. A conditional
permission to assume these roles can be implemented using IAM resource tags.
The iam-sudo utility is an alternative solution which will simulate a defined role, even if
you do not have the permission to assume it. The local implementation is harmless. The remote
implementation provides a backdoor for privilege elevation, so I recommend to deploy it only in
a development environment.
Photo by wildan alfani on Unsplash

Mark van Holsteijn is a senior software systems architect, and CTO of binx.io. He is passionate about removing waste in the software delivery process and keeping things clear and simple.