Operations grimoire/Vault

From Nasqueron Agora

This page currently documents Vault installed on complector as part of rOPS: roles/vault/. It doesn't document Docker or shellserver installation. See Operations grimoire/Eglide/Vault.

Hashicorp Vault allows to manage credentials and is currently used:

  • to provision credentials through Salt
  • to allow applications to fetch credentials
  • to emit certificates for the private network applications
  • to store credentials used by the Nasqueron Operations SIG beings

Vault policies are fully managed through Salt in the operations repository.

Vault has been selected in 2016 for secrets management.

Vault reference

kv engine

The kv engine contains the following paths: 

 * ops/secrets: credentials like passwords, API tokens, private keys deployed to servers - a reference for machines
 * ops/internal: credentials to third-party services internally shared amongst ops - a reference for humans
 * ops/privacy: privacy information

Secrets

Secrets are directly and manually managed in Vault. If we need to run a disaster recovery procedure, we'll roll any secret by defining them again and deploy from rOPS those new secrets.

Privacy data

Personal identity information (PII) shared inside the Nasqueron Operations squad are more convenient to manage as a repository.

Such access to the repository is restricted to commit data, and ensure this data is deployed to the servers needing it. Any other use isn't allowed.

Information to the repository is then published in Vault, and referred in rOPS as Vault credentials.

Note: The ops/privacy path is only intended for PII of current or previous members of the Ops SIG, like IP addresses allowed to connect to restricted resources like the ops VPN. This is not acceptable to put 3rd party information in this repository.

pki engine

Services listening to private IPs use the Vault PKI to generate certificates:

Certificates should be issued short-term and frequently renewed by automated services.

transit engine

Encryption as a service is available for applications under /transit path.

See https://learn.hashicorp.com/tutorials/vault/eaas-transit.

Applications should use a policy restricting to a subpath for that app, not to the whole engine.

Users accounts

Members of the Operations SIG need an access to Vault to create and maintain credentials.

Currently, we use tokens. To generate a token: 

    vault token create -policy=admin

To connect to Vault with a token from a devserver (e.g. WindRiver), you need:

  1. Store the token in $HOME/.vault-token
  2. Define environment variable VAULT_ADDR: export VAULT_ADDR=https://172.27.27.7:8200

Tokens need to be frequently renew (before expiration) with vault renew

Salt configuration

Get a secret through Salt

The information in the ops/secrets and ops/privacy kv engines are integrated in the Salt pillar: 

ext_pillar:
  # Credentials to deploy to servers
  - vault:
       conf: path=ops/secrets
       nesting_key: credentials

  # Personal identity information from Nasqueron Operations SIG members
  - vault:
       conf: path=ops/privacy
       nesting_key: privacy

You can then access to this information in Salt states using the following references:

Access to Salt data
Category Example of path in Vault Example of pillar key in Salt
Secrets ops/secrets/foo pillar['credentials']['foo']
Privacy data ops/privacy/ops-cidr pillar['privacy']['ops-cidr']

The ops/internal paths aren't accessible through Vault.

Policies

Policies are managed in rOPS: pillar/credentials/vault.sls.

For Salt, we deploy secrets to servers, policy management for a node consist to provide the list of secrets a role should have access to, generally in vault_secrets_by_role dictionary.

For other applications:

  • create a policy .hcl in roles/vault/policies/files/
  • declare the policy in the pillar under vault_policies list, it should match the .hcl file (without the extension or the path)
  • run the repository tests to check if both information matches ; to only run that part, use cd _tests && python3 -m unittest discover pillar/credentials/

To add only an application policy, Salt can be deployed like this: 

   salt complector state.sls_id vault_policy_airflow roles/vault/policies

DRP

Any step we need to do to configure the engines should be added to the rOPS: roles/vault/bootstrap unit.

If we need to recover Vault, two solutions are available:

  • restore the storage, to keep data (credentials, certificates including the root CA one)
  • recreate it from scratch through the roles/vault/bootstrap unit

Howto

Allow a new application to authenticate to Vault

Applications need a policy and an authentication method, for example here for an application called yourservice with AppRole with secret_id and role_id:

  1. Create a new policy in operations repository: D3270
  2. Deploy it from Complector with salt complector state.sls_id vault_policy_yourservice roles/vault/policies
  3. Create an approle matching that policy: vault write auth/approle/role/yourservice token_policies="yourservice" token_ttl=1h token_max_ttl=4h
  4. Get role_id with vault read auth/approle/role/yourservice/role-id
  5. Get secret_id with vault write -force auth/approle/role/yourservice/secret-id
  6. If you wish Salt to provision those credentials during your service deployment, probably a good idea to write them too: vault kv put secrets/nasqueron/yourservice/vault username=role-id password=secret-id [to skip if you don't use Salt to provision role-id and secret-id]

Troubleshoot

Connecting to Vault

Access to web UI

Vault isn't listening any public IP, but accept connections from internal network, so you can open a SSH tunnel against complector:8200 and then browse https://localhost:8200:

ssh -L 8200:complector:8200 router-001.nasqueron.org

curl: (60) SSL certificate problem

Curl is compiled against a certificate bundle (curl-config --ca for the path). That bundle doesn't contain the pki_root certificate. We deploy this certificate in /etc/ssl/certs, but curl won't check there if not instructed to.

If you build an immutable environment like a container, add the certificate to the bundle.

On other machines, create a curl configuration file in ~/.curlrc (it doesn't seem currently possible to create a system one) with capath=/etc/ssl/certs.

If you've still the issue after that, check two things:

  • with curl -v, the path should correctly there: * CApath: /etc/ssl/certs in addition to a CAfile.
  • with file /etc/ssl/certs/* | grep -i nasqueron you should have a result ; if not, put the certificate in /usr/local/share/certs and relaunch the tool to link it in /etc/ssl/certs, for example certctl rehash on FreeBSD (needs root access).

Certificates

How to unseal when certificate have expired?

Use the web UI through SSH tunnel.

Renew Vault HTTP certificates

Vault uses TLS certificates managed by its own CA. That can create a chicken/egg issue if certificates have expired to connect to it, as the Vault client strictly requires valid TLS certificated to issue commands to renew certificates.

Connect to the web UI works as your browser is able to bypass certificate requirement. Opening a SSH tunnel with -L 8200:complector:8200 should work for that.

Certificates configuration is available at D2639.

The web UI doesn't offer a comprehensive sets of widgets to generate certificates. You can instead use the Vault Browser CLI to run Vault CLI commands from the web UI. The signing operation can be done through endpoint configuration UI.

Based on current D2639 state (2023-01), you can do the following:

  • Has intermediate authority certificate expired?
    • Web console: write -format=json pki_vault/intermediate/generate/internal common_name="nasqueron.drake Intermediate Authority"
    • At https://localhost:8200/ui/vault/settings/secrets/configure/pki_root/cert you can use the sign intermediate button to sign it, don't forget to fill the following information:
      • CSR, pick from console the .data.csr output from JSON response
      • Format: pem_bundle
      • TTL: 100 days
      • Common name: "nasqueron.drake Intermediate Authority"
      • Organization: Nasqueron
      • Organization unit: Nasqueron Operations SIG
      • Country: BE
    • Follow the warning instruction: copy the certificate bundle, and in pki_vault, use Set signed intermediate to set the certificate as intermediate one. That will create a new issuer, fetch the name in "issuers".
    • Save also the certificate bundle to /usr/local/share/certs/nasqueron-vault-intermediate.crt
    • Edit the nasqueron-drake role to use your new issuer name
  • Has complector.nasqueron.drake certificate has expired?
    • Web console: write -format=json pki_vault/issue/nasqueron-drake common_name=complector.nasqueron.drake ttl=2160h ip_sans=127.0.0.1,172.27.27.7, save the output in /usr/local/etc/certificates/vault files:
      • .data.certificate to certificate.pem
      • .data.issuing_ca to ca.pem
      • .data.private_key to private.key (careful how you replace the \n, if you use Python REPL, do it on Complector itself and get rid of the history with import readline ; readline.clear_history())
    • Prepare the fullchain bundle with cat certificate.pem ca.pem > fullchain.pem
    • Restart vault with /usr/local/etc/rc.d/vault restart, UI should pick new certificate and you can know directly use the vault command. Vault configuration currently uses fullchain.pem and private.key.

CSR uses line1\nline2\nline3 format in JSON. You can use a Python REPL with print("...") to print the CSR in several lines. CSR isn't confidential information, key is protected in Vault and never exposed.

jq. Another tool that can help is jq. For example, cat certificate.json | jq -r .data.certificate > /usr/local/etc/certificates/vault/certificate.pem will save it in multiline format.

If you get an issue about private key missing at Set signed intermediate step, don't forget you're setting an intermediate certificate for pki_vault, and not for pki_root, set that here: https://localhost:8200/ui/vault/settings/secrets/configure/pki_vault/cert

Paths where to store certificates matter:

  • pki_root certificates are stored in /usr/local/share/certs
  • pki_vault certificates are stored in /usr/local/etc/certificates/vault

Ensure `private.key` belongs to user `vault` and receives a correct chmod like 400.

Services known to break when a certificate is expired

  • Salt, but that's ad-hoc and explicit
  • Sentry containers: sudo docker ps -a | grep sentry | grep -v "Up "

Salt

credentials.vault_policy_present returns Bad Request

Vault returns a 400 error if the policy format is incorrect.

Case 1. The state ID starts by vault_policy_. You manually added a file to roles/vault/policies/file: this error means Vault API founds a syntax error in it. You can use https://github.com/mschuchard/linter-vault-policy if you use Pulsar (ex Atom) to lint those files.

Case 2. The state ID starts by anything else. The policy has been generated through _modules/credentials.py.

Check if the pillar values added are correct:

  • if not, add a test in _tests/pillar/credentials/test_vault.py to detect the case
  • if yes, there is an issue with the Python module code to fix

Example of Salt output. 

----------
          ID: vault_policy_admin
    Function: credentials.vault_policy_present
        Name: admin
      Result: False
     Comment: Failed to change policy: Bad Request
     Started: 13:45:14.162421
    Duration: 76.168 ms
     Changes:

Current status

  • Secrets in pillar/credentials/zr.sls have been migrated to Vault under ops/secrets/<key>
Retrieved from "https://agora.nasqueron.org/index.php?title=Operations_grimoire/Vault&oldid=1498"