SSM-Scala is a command-line tool, written in Scala, for executing commands on EC2 servers using AWS's EC2 Run command. It provides the user with:
- standard
sshaccess using short lived RSA keys - an alternative to
sshfor running commands on the target
Both modes apply to servers in AWS accounts to which you have IAM access.
Instructions for using SSM Scala in your own project can be found below.
If you have Homebrew installed and want to install ssm, do
brew install guardian/devtools/ssm
and for an upgrade do
brew upgrade ssm
Otherwise, fetch the most recently released version of the program from the Github releases page and make sure it is executable (chmod +x ssm). You may then want to put it somewhere in your PATH.
The readme is quite detailed (and shows how to do many more things than what will be shown in this section) but you are probably reading it because you just want to ssh to a box. Here is what you need to do:
-
Install ssm. How to do so was explained in the previous section.
-
Ensure that you have the Janus credentials of the account you want to work with. We are going to assume
frontendin this section for the examples. -
Identify the instance number of the box you want to reach. It can be found in the AWS developer console. Instance numbers look like this
i-00032c76140bc9140. -
At your console type
ssm ssh -i i-00032c76140bc9140 -p frontendand more generally
ssm ssh -i <instance-id> -p <accountName> ssm ssh -i <instance-id> -p <accountName> -
And that's it! If all went well you have been ssh'ed to the box.
If you get an error about Futures timed out after 25 seconds, then the SSM permissions may not be right, or you might need to recycle the instance since adding the permissions.
If the disk on which the keyfile is stored is full, then ssm-scala cannot add the public key identity prior to logging in to the box. This is often found to be the case, and also can apparently cause the AWS SSM agent to stop.
One potential workaround for this is rebooting the box using the EC2 console (may clear down logs, for example).
The automatically generated help section for ssm is
Usage: ssm [cmd|repl|ssh|scp] [options] <args>...
-p, --profile <value> The AWS profile name to use for authenticating this execution
-i, --instances <value> Specify the instance ID(s) on which the specified command(s) should execute
-t, --tags <value> Search for instances by tag. If you provide less than 3 tags assumed order is app,stage,stack. e.g. '--tags riff-raff,prod' or '--tags grafana' Upper/lowercase variations will be tried.
-r, --region <value> AWS region name (defaults to eu-west-1)
--verbose enable more verbose logging
--use-default-credentials-provider
Use the default AWS credentials provider chain rather than profile credentials. This option is required when running within AWS itself.
Command: cmd [options]
Execute a single (bash) command, or a file containing bash commands
-u, --user <value> Execute command on remote host as this user (default: ubuntu)
-c, --cmd <value> A bash command to execute
-f, --file <value> A file containing bash commands to execute
Command: repl
Run SSM in interactive/repl mode
Command: ssh [options]
Create and upload a temporary ssh key
-u, --user <value> Connect to remote host as this user (default: ubuntu)
--port <value> Connect to remote host on this port
--newest Selects the newest instance if more than one instance was specified
--oldest Selects the oldest instance if more than one instance was specified
--private Use private IP address (must be routable via VPN Gateway)
--raw Unix pipe-able ssh connection string. Note: disables automatic execution. You must use 'eval' to execute this due to nested quoting
-x, --execute [Deprecated - new default behaviour] Makes ssm behave like a single command (eg: `--raw` with automatic piping to the shell)
-d, --dryrun Generate SSH command but do not execute (previous default behaviour)
-A, --agent Use the local ssh agent to register the private key (and do not use -i); only bastion connections
-a, --no-agent Do not use the local ssh agent
-b, --bastion <value> Connect through the given bastion specified by its instance id; implies -A (use agent) unless followed by -a.
-B, --bastion-tags <value>
Connect through the given bastion identified by its tags; implies -a (use agent) unless followed by -A.
--bastion-port <value> Connect through the given bastion at a given port.
--bastion-user <value> Connect to bastion as this user (default: ubuntu).
--host-key-alg-preference <value>
The preferred host key algorithms, can be specified multiple times - last is preferred (default: ecdsa-sha2-nistp256, ssh-rsa)
--ssm-tunnel [deprecated]
--no-ssm-proxy Do not connect to the host proxying via AWS Systems Manager - go direct to port 22. Useful for instances running old versions of systems manager (< 2.3.672.0)
--tunnel <value> Forward traffic from the given local port to the given host and port on the remote side. Accepts the format `localPort:host:remotePort`, e.g. --tunnel 5000:a.remote.host.com:5000
--rds-tunnel <value> Forward traffic from a given local port to a RDS database specified by tags. Accepts the format `localPort:tags`, where `tags` is a comma-separated list of tag values, e.g. --rds-tunnel 5000:app,stack,stage
Command: scp [options] [:]<sourceFile>... [:]<targetFile>...
Secure Copy
-u, --user <value> Connect to remote host as this user (default: ubuntu)
--port <value> Connect to remote host on this port
--newest Selects the newest instance if more than one instance was specified
--oldest Selects the oldest instance if more than one instance was specified
--private Use private IP address (must be routable via VPN Gateway)
--raw Unix pipe-able scp connection string
-x, --execute [Deprecated - new default behaviour] Makes ssm behave like a single command (eg: `--raw` with automatic piping to the shell)
-d, --dryrun Generate SCP command but do not execute (previous default behaviour)
--ssm-tunnel [deprecated]
--no-ssm-proxy Do not connect to the host proxying via AWS Systems Manager - go direct to port 22. Useful for instances running old versions of systems manager (< 2.3.672.0)
[:]<sourceFile>... Source file for the scp sub command. See README for details
[:]<targetFile>... Target file for the scp sub command. See README for details
There are two mandatory configuration items.
To specify your AWS profile (for more information see AWS profiles), either of:
--profile- AWS_PROFILE environment variable
To target the command, either of:
-i, where you specify one or more instance ids, or-t, where you specify the app name, the stack and the stage.
When accessing to an instance the user is greeted with a message of the form
This instance should be considered tainted.
It was accessed by 1234567890:alice.smith at Fri Apr 27 08:36:58 BST 2018
This message highlights the fact that access is being logged and that the next person will see that the current user has been there. The current wording of considered tainted highlights the fact that the user has no idea what has happened during previous ssh sessions and raises awareness of the implications of accessing a box.
This is the result of having too many keys in your agent and exceeding the servers configured authentication attempts. This is fixed as of 0.9.7 as we disable the use of the agent using IdentitiesOnly.
If you still see "Too many authentication failures" then please raise an issue. You can work around it by running ssh-add -D to remove all keys from your agent.
If you need to add extra parameters to the SSH command then you can use --raw. In it's simplest form the following are equivalent:
ssm ssh -i i-0123456789abcdef0 -p composerand
eval $(ssm ssh -i i-0123456789abcdef0 -p composer --raw)This helps to undertake actions such as construct tunnels. For example to access a remote postgres server:
eval $(ssm ssh -i i-0123456789abcdef0 -p composer --raw) -L 5432:my-postgres-server-hostname:5432Note the use of eval in these examples - this is required in order to correctly parse the nested quotes that are output as part of the raw command. If you don't use eval then you are likely to see an error message such as ssh: Could not resolve hostname yes": nodename nor servname provided, or not known.
ssm needs to be told which instances should execute the provided command(s). You can do this by specifying instance IDs, or by specifying App, Stack, and Stage tags.
# by instance ids
--instances i-0123456,i-9876543
-i i-0123456,i-9876543
# by tag
--tags <app>,<stage>,<stack>
-t <app>,<stage>,<stack>
If you provide tags, ssm will search for running instances that are have those tags.
Examples of using cmd are
./ssm cmd -c date --profile security -t security-hq,security,PROD
or
export AWS_PROFILE=security
./ssm cmd -c date -t security-hq,security,PROD
where the date command will be ran on all matching instances.
An example of using repl is:
./ssm repl --profile <aws-profile> -t security-hq,security,PROD
The REPL mode causes ssm to generate a list of instances and then wait for commands to be specified. Each command will be executed on all instances and the user can select the instance to display.
An example of using ssh command is:
./ssm ssh --profile <aws-profile> -t security-hq,security,PROD
This causes ssm to generate a temporary ssh key, and install the public key on a specific instance. It will then output the command to ssh directly to that instance. The instance must already have appropriate security groups.
The target for the ssh command will be the public IP address if there is one, otherwise the private IP address. The --private flag overrides this behavior and defaults to the private IP address.
Note that if the argument -t <app>,<stack>,<stage> resolves to more than one instance, the command will stop with an error message. You can circumvent this behaviour and instruct ssm to proceed with one single instance using the command line flags --oldest and --newest, which select either the oldest or newest instances.
This flag allows for a pipe-able ssh connection string. For instance
ssm ssh --profile security -t security-hq,security,PROD --newest --raw | xargs -0 -o bash -c
Will automatically ssh you to the newest instance running security-hq. Note that you still have to manually accept the new ECDSA key fingerprint.
Generate SSH command but do not execute (previous default behaviour)
ssm ssh --profile security -t security-hq,security,PROD --newest --dryrun
Example output:
========= i-0566a4df63c0c35bb =========
# Dryrun mode. The command below will remain valid for 30 seconds:
ssh -o "IdentitiesOnly yes" -o "UserKnownHostsFile ...
DEPRECATED - flag is now the default behaviour. This flag makes ssm behave like ssh. The raw output is automatically piped to xargs -0 -o bash -c. You would then do
ssm ssh --profile security -t security-hq,security,PROD --newest --execute
instead of the example given in the previous --raw section.
This flag forwards traffic from a local port through the instance to the specified hostname and port. For example,
ssm ssh --profile security -t security-hq,security,PROD --newest --tunnel 5000:example.com:6000
would forward all traffic on your machine through the remote instance to example.com:6000.
Similar to tunnel, this flag forwards traffic from a local port to an AWS RDS database specified by the given tags. For example,
ssm ssh --profile security -t security-hq,security,PROD --newest --rds-tunnel 5000:example-db,security,CODE
would try to find a single RDS instance with the tags example-db,security,CODE, and forward traffic from port 5000 to that RDS instance via the remote instance.
By default, SSM proxies your connection via AWS systems manager, which saves you from opening up port 22, connecting to
the VPN, or using bastion hosts. This requires a recent version of systems manager to be runnning on your machine and
the target machine. You can still connect the old way via port 22 using the flag --no-ssm-proxy
It is strongly encouraged to connect using the default SSM tunnel behaviour. To get this working you'll need to do the following stuff:
Update the permissions of your instances so that they are allowed to do these things:
- ec2messages:AcknowledgeMessage
- ec2messages:DeleteMessage
- ec2messages:FailMessage
- ec2messages:GetEndpoint
- ec2messages:GetMessages
- ec2messages:SendReply
- ssm:UpdateInstanceInformation
- ssm:ListInstanceAssociations
- ssm:DescribeInstanceProperties
- ssm:DescribeDocumentParameters
- ssmmessages:CreateControlChannel
- ssmmessages:CreateDataChannel
- ssmmessages:OpenControlChannel
- ssmmessages:OpenDataChannel
See here for an example complete policy.
You'll also need to ensure you're using a recent AMI that has at least version 2.3.672.0 of systems manager - this is now in our base images so using a recent amigo AMI should do the job.
Once these permissions are added, the ssm-agent service running on the boxes will need to be restarted before connecting. This will happen as boxes are cycled – e.g. by redeploying your app – or you can restart an agent manually with sudo snap restart amazon-ssm-agent.amazon-ssm-agent.
Upgrade your local version of ssm and awscli:
brew upgrade ssm
brew upgrade awscli
You'll also need to install the systems manager plugin on your machine:
brew cask install session-manager-plugin
You can then SSH using SSM with the default arguments:
ssm ssh -i i-0937fe9baa578095b -p deployTools
(Useful tip - you can find the instance id using prism, e.g. prism -f instanceName amigo)
Once you've confirmed this is working you can remove any security group rules allowing access on port 22.
Check out the original PR: #111 for further details on how this works.
Bastion are proxy servers used as entry point to private networks and ssm scala supports their use.
You may not need a bastion server at all! Prefer to use an SSM tunnel (see above) where possible.
In this example we assume that you have a bastion with a public IP address (even though the bastion Ingress rules may restrict it to some IP ranges), identified by aws instance id i-bastion12345, and an application server, on a private network with private IP address, and with instance id i-application-12345, you would then use ssm to connect to it using
ssm ssh --profile <profile-name> --bastion i-bastion12345 --bastion-port 2022 -i i-application-12345
The outcome of this command is a one-liner of the form
ssh -A -i /path/to/private/key-file [email protected] -t -t ssh [email protected];
You can specify a port that the bastion runs ssh on, with the option --bastion-port <port number>, example
ssm ssh --profile <profile-name> --bastion i-bastion12345 --bastion-port 2345 -i i-application-12345
In the current version of bastion support you will need to specify the bastion using its aws instance id, but you can refer to the application instance using the tag system as in
ssm ssh --profile <profile-name> --bastion i-bastion12345 --bastion-port 2022 --tags app,stage,stack
together, if the tags may resolve to more than one instance, the --oldest and --newest flags
ssm ssh --profile <profile-name> --bastion i-bastion12345 --bastion-port 2022 --tags app,stage,stack --newest
If you do not know the id of the current bastion, but it is tagged correctly, it is also possible to use:
ssm ssh --profile <profile-name> --bastion-tags <app,stage,stack> --bastion-port 2022 -i i-application-12345
This will respect any --newest / --oldest switches, although it is anticipated that there will usually only be one bastion. It will always use the public IP address of the bastion.
It is possible to specify the user used for connecting to the bastion, this is done with the --bastion-user <value> command line argument.
When using the standard ssh command, the --private flag can be used to indicate that the private IP of the target instance should be used for the connection. In the case of bastion connection the target instance is assumed to always be reacheable through a private IP and this flag indicates whether the private IP of the bastion should be used.
There's been occurences of bastions connections strings of the form
ssh -A -i /path/to/temp/private/key -t -t ubuntu@bastion-hostname \
-t -t ssh -t -t ubuntu@target-ip-address;
not working, because the private file was not found for the second ssh connection, leading to a "Permission denied (publickey)" error message.
When this happens the user can use the -a, --agent flag that performs a registration of the private key at the local ssh agent. With this flag, ssm command
ssm ssh --profile <account-name> --bastion <instanceId1> \
-i <instanceId2> --agent
returns
ssh-add /path/to/temp/private/key && \
ssh -A ubuntu@bastion-hostname \
-t -t ssh ubuntu@target-ip-address;
ssm support the scp sub command for the secure transfer of files and directories.
An example of usage is
./ssm scp -p account -t app,stage,stack /path/to/file1 :/path/to/file1
Which outputs
# simplified version
scp -i /path/to/identity/file.tmp /path/to/file1 [email protected]:/path/to/file2;
Otherwise
./ssm scp -p account -t app,stage,stack :/path/to/file1 /path/to/file2
outputs
# simplified version
scp -i /path/to/identity/file.tmp [email protected]:/path/to/file1 /path/to/file2 ;
The convention is: the first (left hand side) file is always the source and the second (right hand side) is always the target and the colon, indicates which one is on the remote server.
During development, the program can be run using sbt, either from an sbt shell or from the CLI in that project.
$ sbt "run cmd -c pwd --instances i-0123456 --profile xxx --region xxx"
sbt:ssm-scala> run cmd -c pwd --instances i-0123456 --profile xxx --region xxx
However, sbt traps the program exit so in REPL mode you may find it easier to create and run an executable instead, for this just run
./generate-executable.sh
The result of this script is an executable called ssm in the target folder. If you are using a non unix operating system, run sbt assembly as you would normally do and then run the ssm.jar file using
java -jar <path-to-jar>/ssm.jar [arguments]
To release a new version of ssm perform the two following tasks:
-
Update the version number in
build.sbt -
Generate a new executable. Run the following at the top of the repository
./generate-executable.shNote that this script generates the tar.gz file needed for the github release as well as outputting the sha256 hash of that file needed for the homebrew-devtools' update.
-
Increase the version number accordingly and release a new tag at ssm-scala releases. Upload the raw executable (file:
ssm) as well as a tar.gz version (file:ssm.tar.gz). -
Make a PR to https://github.com/guardian/homebrew-devtools/blob/master/Formula/ssm.rb to update the new version's details.
To use ssm-scala against the instances of your project, the following needs to happen:
-
Add permissions with a policy like:
ExampleAppSSMRunCommandPolicy: Type: AWS::IAM::Policy Properties: PolicyName: example-app-ssm-run-command-policy PolicyDocument: Statement: # minimal policy to allow to (only) run commands via ssm - Effect: Allow Resource: "*" Action: - ec2messages:AcknowledgeMessage - ec2messages:DeleteMessage - ec2messages:FailMessage - ec2messages:GetEndpoint - ec2messages:GetMessages - ec2messages:SendReply - ssm:UpdateInstanceInformation - ssm:ListInstanceAssociations - ssm:DescribeInstanceProperties - ssm:DescribeDocumentParameters - ssmmessages:CreateControlChannel - ssmmessages:CreateDataChannel - ssmmessages:OpenControlChannel - ssmmessages:OpenDataChannel Roles: - !Ref ExampleAppInstanceRole
Example stolen from the Security-HQ cloudformation file.
-
Download the executable from the project release page. Instructions on usage can be found in the above sections.
Note: SSM needs the target server to have outbound port 443 (ssm-agent's communication with AWS's SSM and EC2 Messages endpoints).
##License
Copyright (c) 2018 Guardian News & Media. Available under the Apache License.