open-vault/website/content/docs/agent-and-proxy/autoauth/index.mdx
hc-github-team-secure-vault-core e147778434
backport of commit f54bc0227d56ad01aa5fe88ff4288cb1ccf3ce4d (#21939)
Co-authored-by: Violet Hynes <violet.hynes@hashicorp.com>
2023-07-19 14:10:08 +00:00

348 lines
12 KiB
Plaintext

---
layout: docs
page_title: Vault Agent and Vault Proxy Auto-Auth
description: |-
Vault Agent and Vault Proxy's Auto-Auth functionality allows easy and automatic
authentication to Vault in a variety of environments.
---
# Vault Agent and Vault Proxy Auto-Auth
The Auto-Auth functionality of Vault Agent and Vault Proxy allow for
easy authentication in a wide variety of environments.
## Functionality
Auto-Auth consists of two parts: a Method, which is the authentication method
that should be used in the current environment; and any number of Sinks, which
are locations where the agent should write a token any time the current token
value has changed.
When Vault Agent or Vault Proxy are started with Auto-Auth enabled, it will attempt to acquire a
Vault token using the configured Method. On failure, it will exponentially back
off and then retry. On success, unless the auth method is configured to wrap
the tokens, it will keep the resulting token renewed until renewal is no longer
allowed or fails, at which point it will attempt to reauthenticate.
Every time an authentication is successful, the token is written to the
configured Sinks, subject to their configuration.
## Advanced functionality
Sinks support some advanced features, including the ability for the written
values to be encrypted or
[response-wrapped](/vault/docs/concepts/response-wrapping).
Both mechanisms can be used concurrently; in this case, the value will be
response-wrapped, then encrypted.
### Response-Wrapping tokens
There are two ways that tokens can be response-wrapped:
1. By the auth method. This allows the end client to introspect the
`creation_path` of the token, helping prevent Man-In-The-Middle (MITM)
attacks. However, because auto-auth cannot then unwrap the token and rewrap
it without modifying the `creation_path`, we are not able to renew the
token; it is up to the end client to renew the token. Agent and Proxy both
stay daemonized in this mode since some auth methods allow for reauthentication
on certain events.
2. By any of the token sinks. Because more than one sink can be configured, the
token must be wrapped after it is fetched, rather than wrapped by Vault as
it's being returned. As a result, the `creation_path` will always be
`sys/wrapping/wrap`, and validation of this field cannot be used as
protection against MITM attacks. However, this mode allows auto-auth to keep
the token renewed for the end client and automatically reauthenticate when
it expires.
### Encrypting tokens
~> Support for encrypted tokens is experimental; if input/output formats
change, we will make every effort to provide backwards compatibility.
Tokens can be encrypted, using a Diffie-Hellman exchange to generate an
ephemeral key. In this mechanism, the client receiving the token writes a
generated public key to a file. The sink responsible for writing the token to
that client looks for this public key and uses it to compute a shared secret
key, which is then used to encrypt the token via AES-GCM. The nonce, encrypted
payload, and the sink's public key are then written to the output file, where
the client can compute the shared secret and decrypt the token value.
~> NOTE: Token encryption is not a protection against MITM attacks! The purpose
of this feature is for forward-secrecy and coverage against bare token values
being persisted. A MITM that can write to the sink's output and/or client
public-key input files could attack this exchange. Using TLS to protect the
transit of tokens is highly recommended.
To help mitigate MITM attacks, additional authenticated data (AAD) can be
provided to Agent and Proxy. This data is written as part of the AES-GCM tag and must
match on both Agent and Proxy and the client. This of course means that protecting
this AAD becomes important, but it provides another layer for an attacker to
have to overcome. For instance, if the attacker has access to the file system
where the token is being written, but not to read configuration or read
environment variables, this AAD can be generated and passed to Agent or Proxy and
the client in ways that would be difficult for the attacker to find.
When using AAD, it is always a good idea for this to be as fresh as possible;
generate a value and pass it to your client and Agent or Proxy on startup. Additionally,
Agent and Proxy a Trust On First Use model; after it finds a generated public key,
it will reuse that public key instead of looking for new values that have been
written.
If writing a client that uses this feature, it will likely be helpful to look
at the
[dhutil](https://github.com/hashicorp/vault/blob/main/helper/dhutil/dhutil.go)
library. This shows the expected format of the public key input and envelope
output formats.
## Configuration
The top level `auto_auth` block has two configuration entries:
- `method` `(object: required)` - Configuration for the method
- `sinks` `(array of objects: optional)` - Configuration for the sinks
### Configuration (Method)
~> Auto-auth does not support using tokens with a limited number of uses. Auto-auth
does not track the number of uses remaining, and may allow the token to
expire before attempting to renew it. For example, if using AppRole auto-auth,
you must use 0 (meaning unlimited) as the value for
[`token_num_uses`](/vault/api-docs/auth/approle#token_num_uses).
These are common configuration values that live within the `method` block:
- `type` `(string: required)` - The type of the method to use, e.g. `aws`,
`gcp`, `azure`, etc. _Note_: when using HCL this can be used as the key for
the block, e.g. `method "aws" {...}`.
- `mount_path` `(string: optional)` - The mount path of the method. If not
specified, defaults to a value of `auth/<method type>`.
- `namespace` `(string: optional)` - Namespace in which the mount lives.
The order of precedence is: this setting lowest, followed by the
environment variable `VAULT_NAMESPACE`, and then the highest precedence
command-line option `-namespace`.
If none of these are specified, defaults to the root namespace.
Note that because sink response wrapping and templating are also based
on the client created by auto-auth, they use the same namespace.
- `wrap_ttl` `(string or integer: optional)` - If specified, the written token
will be response-wrapped by auto-auth. This is more secure than wrapping by
sinks, but does not allow the auto-auth to keep the token renewed or
automatically reauthenticate when it expires. Rather than a simple string,
the written value will be a JSON-encoded
[SecretWrapInfo](https://godoc.org/github.com/hashicorp/vault/api#SecretWrapInfo)
structure. Uses [duration format strings](/vault/docs/concepts/duration-format).
- `min_backoff` `(string or integer: "1s")` - The minimum backoff time auto-auth
will delay before retrying after a failed auth attempt. The backoff will start
at the configured value and double (with some randomness) after successive
failures, capped by `max_backoff.` If Agent templating is being used, this
value is also used as the min backoff time for the templating server.
Uses [duration format strings](/vault/docs/concepts/duration-format).
- `max_backoff` `(string or integer: "5m")` - The maximum time Agent will delay
before retrying after a failed auth attempt. The backoff will start at
`min_backoff` and double (with some randomness) after successive failures,
capped by `max_backoff.` If Agent templating is being used, this value is also
used as the max backoff time for the templating server. `max_backoff` is the
duration between retries, and **not** the duration that retries will be
performed before giving up. Uses [duration format strings](/vault/docs/concepts/duration-format).
- `exit_on_err` `(bool: false)` - When set to true, Vault Agent and Vault Proxy
will exit if any errors occur during authentication. This configurable only affects login
attempts for new tokens (either initial or expired tokens) and will not exit for errors on
valid token renewals.
- `config` `(object: required)` - Configuration of the method itself. See the
sidebar for information about each method.
### Configuration (Sinks)
These configuration values are common to all Sinks:
- `type` `(string: required)` - The type of the method to use, e.g. `file`.
_Note_: when using HCL this can be used as the key for the block, e.g. `sink "file" {...}`.
- `wrap_ttl` `(string or integer: optional)` - If specified, the written token
will be response-wrapped by the sink. This is less secure than wrapping by
the method, but allows auto-auth to keep the token renewed and automatically
reauthenticate when it expires. Rather than a simple string, the written
value will be a JSON-encoded
[SecretWrapInfo](https://godoc.org/github.com/hashicorp/vault/api#SecretWrapInfo)
structure. Uses [duration format strings](/vault/docs/concepts/duration-format).
- `dh_type` `(string: optional)` - If specified, the type of Diffie-Hellman exchange to
perform, meaning, which ciphers and/or curves. Currently only `curve25519` is
supported.
- `dh_path` `(string: required if dh_type is set)` - The path from which the
auto-auth should read the client's initial parameters (e.g. curve25519 public
key).
- `derive_key` `(bool: false)` - If specified, the final encryption key is
calculated by using HKDF-SHA256 to derive a key from the calculated shared
secret and the two public keys for enhanced security. This is recommended
if backward compatibility isn't a concern.
- `aad` `(string: optional)` - If specified, additional authenticated data to
use with the AES-GCM encryption of the token. Can be any string, including
serialized data.
- `aad_env_var` `(string: optional)` - If specified, AAD will be read from the
given environment variable rather than a value in the configuration file.
- `config` `(object: required)` - Configuration of the sink itself. See the
sidebar for information about each sink.
### Auto auth examples
Auto-Auth configuration objects take two separate forms when specified in HCL
and JSON. The following examples are meant to clarify the differences between
the two formats.
#### Sinks (HCL format)
The HCL format may define any number of sink objects with an optional wrapping
`sinks {...}` object.
~> Note: The [corresponding JSON format](#sinks-json-format) _must_ specify a
`"sinks" : [...]` array to encapsulate all `sink` JSON objects.
```hcl
// Other Vault Agent or Vault Proxy configuration blocks
// ...
auto_auth {
method {
type = "approle"
config = {
role_id_file_path = "/etc/vault/roleid"
secret_id_file_path = "/etc/vault/secretid"
}
}
sinks {
sink {
type = "file"
config = {
path = "/tmp/file-foo"
}
}
}
}
```
The following valid HCL omits the wrapping `sinks` object while specifying
multiple sinks.
```hcl
// Other Vault Agent or Vault Proxy configuration blocks
// ...
auto_auth {
method {
type = "approle"
config = {
role_id_file_path = "/etc/vault/roleid"
secret_id_file_path = "/etc/vault/secretid"
}
}
sink {
type = "file"
config = {
path = "/tmp/file-foo"
}
}
sink {
type = "file"
config = {
path = "/tmp/file-bar"
}
}
}
```
#### Sinks (JSON format)
The following JSON configuration illustrates the need for a `sinks: [...]` array
wrapping any number of `sink` objects.
```hcl
{
"auto_auth" : {
"method" : [
{
type = "approle"
config = {
role_id_file_path = "/etc/vault/roleid"
secret_id_file_path = "/etc/vault/secretid"
}
}
],
"sinks" : [
{
"sink" : {
type = "file"
config = {
path = "/tmp/file-foo"
}
}
}
]
}
}
```
Multiple sinks are defined by appending more `sink` objects within the `sinks`
array:
```hcl
{
"auto_auth" : {
"method" : [
{
type = "approle"
config = {
role_id_file_path = "/etc/vault/roleid"
secret_id_file_path = "/etc/vault/secretid"
}
}
],
"sinks" : [
{
"sink" : {
type = "file"
config = {
path = "/tmp/file-foo"
}
}
},
{
"sink" : {
type = "file"
config = {
path = "/tmp/file-bar"
}
}
}
]
}
}
```