Hardening Guide

Application configuration

The subsections below cover hardening related to vaultwarden itself.

Disable registration and (optionally) invitations

By default, vaultwarden allows any anonymous user to register new accounts on the server without first being invited. While this is not necessary if you have access to the admin page, this is useful if you're the first user on the server and it's recommended that you disable it in the admin panel (if the admin panel is enabled) or with the environment variable to prevent attackers from creating accounts on your vaultwarden server.

vaultwarden also allows registered users to invite other new users to create accounts on the server and join their organizations. This does not pose an immediate risk (as long as you trust your users), but it can be disabled in the admin panel or with the environment variable.

Disable password hint display

vaultwarden displays password hints on the login page to accommodate small/local deployments that do not have SMTP configured, which could be abused by an attacker to facilitate password-guessing attacks against users on the server. This can be disabled in the admin panel by unchecking the Show password hints option or with the environment variable.

HTTPS / TLS configuration

The subsections below cover hardening related to HTTPS / TLS.

Strict SNI

SNI is how a web browser requests that an HTTPS server provide the SSL/TLS certificate for a particular site (e.g., bitwarden.example.com). Suppose bitwarden.example.com has the IP address 1.2.3.4. Ideally, you want your instance to only be accessible via https://bitwarden.example.com, and not https://1.2.3.4. This is because IP addresses are continually scanned for various reasons, and your vaultwarden instance becomes a more obvious target if it can be detected this way. For example, a simple Shodan search reveals a number of Bitwarden instances that are accessible by IP address.

Reverse proxying

In general, you should avoid enabling HTTPS via vaultwarden's built-in Rocket TLS support, especially if your instance is publicly accessible. Rocket itself lists the following warning:

Rocket's built-in TLS is not considered ready for production use. It is intended for development use only.

For example, Rocket TLS doesn't support strict SNI or ECC certs (only RSA).

See Proxy example for some sample reverse proxy configurations.

Access logs contain access_token parameter

A WSS connection is established for notifications by invoking a GET request with a JWT key.

Example GET request:

/notifications/hub?access_token=[this part is always the same].eyJuYmYi[redacted]sImV4cCI6MTcxNzc1NzQ1OCwiaXN[redacted]M6Ly92YXVsdC5zZWMuYXJwYXxsb2dpbiIsInN1YiI6ImY5YmVhN[redacted]tNGJjNS05MDY2LTQ3NjFlZmY4ND[redacted]sInByZW1pdW0iOnRydWU[redacted]JjaXBoZXIiLCJlbWFpbCI6ImNpc[redacted]ljdSIsImVtYWlsX3ZlcmlmaWVkIjp0cnVlLCJzc3RhbXAiOiJlZjM3[redacted]MjctODE2OS1hZTQ3NmFjNDc4MGQiLCJkZX[redacted]02ZTk3LTQ2N2M[redacted]jM3NmEiLCJzY29wZSI6WyJhcG[redacted]5lX2FjY2VzcyJdLCJhbXIiOlsiQXBwbGljY[redacted]hGDeCNdjTs1cOL2fV_OR96Sey-gA5eRa8OCGNgCrDeyYAPyk[redacted]BkQGwjEhD7fcWILxRYqQ7W6rkC2o[redacted]LB_nztpAgeRUbsPgsd3RNTWJDKdlH8aMf1[redacted]vB_doENJPeyaeMuEG85KqpAN2A[redacted]GeeCztxmQIe21PMtBG-SAgGeI[redacted]X_9mmyv0nISHBuHjhQ_km[redacted]VCLoFneb-MEzN[redacted]T8VcXSKhGXpwJUx8j1[redacted]k_nH27vrD2Dg

If your reverse proxy is configured to save access logs, or the access logs are shipped to an external service such as Prometheus + Promtail, it is recommended to redact the value of the access_token parameter on the external log store, or directly on your reverse proxy of choice, if supported.

No other data is ever send via GET request, encrypted or unencrypted.

Be advised that internal vaultwarden logging strips query to 30 chars, hence access_token is truncated, meaning if not using reverse proxy, you shall be safe from this by default.

Docker configuration

The subsections below cover hardening related to Docker.

Run as a non-root user

The Vaultwarden Docker image is configured to run the container process as the root user by default. This allows Vaultwarden to read/write any data bind-mounted into the container without permission issues, even if that data is owned by another user (e.g., your user account on the Docker host).

The default configuration provides a good balance of security and usability -- running as root within an unprivileged Docker container provides a reasonable level of isolation on its own, while also making setup easier for users who aren't necessarily well-versed in how to manage ownership/permissions on Linux. However, as a general policy, it's better security-wise to run processes with the minimum privileges required; this is somewhat less of a concern with programs written in a memory-safe language like Rust, but note that Vaultwarden does also use some library code written in C (SQLite, OpenSSL, MySQL, PostgreSQL, etc.).

To run the container process (vaultwarden) as a non-root user (uid/gid 1000) in Docker:

docker run -u 1000:1000 [...other args...] vaultwarden/server:latest

To do similarly in docker-compose:

services:
  vaultwarden:
    image: vaultwarden/server:latest
    container_name: bitwarden
    user: 1000:1000
    ... other configuration ...

The default user in many Linux distros has uid/gid 1000 (run the id command to verify), so this is a good value to use if you prefer to be able to easily access your Vaultwarden data without changing to another user, but you can adjust the uid/gid as needed. Note that you'll most likely need to specify a numeric uid/gid, because the Vaultwarden container doesn't share the same mapping of user/group names to uid/gid (e.g., compare the /etc/passwd and /etc/group files in the container to the ones on the Docker host).

The Vaultwarden Docker images are set up such that the vaultwarden executable binds to port 80, which works fine since it runs as root by default. However, a non-root process is normally unable to bind to a privileged port (i.e., a port below 1024). As of version 20.10.0 (see moby/moby#41030), Docker specially configures its containers so that non-root processes are allowed to bind to privileged ports by default. For earlier versions of Docker, or other container runtimes without this special behavior, the Vaultwarden Docker images also sets the cap_net_bind_service capability on the vaultwarden executable, which is another way to allow an executable to bind to privileged ports when running as a non-root user.

Mounting data into the container

Generally, only data that vaultwarden needs to operate properly should be mounted into the vaultwarden container (typically, this is just your data directory, and maybe a directory containing SSL/TLS certs and private keys). For example, don't mount your entire home directory, /var/run/docker.sock, etc. unless you have a specific reason and know what you're doing.

Also, if you don't expect vaultwarden to modify the data you're mounting in (e.g., certs), then mount it read-only by adding :ro to the volume specification (for example, docker run -v /home/username/bitwarden-ssl:/ssl:ro).

Miscellaneous

Brute-force mitigation

When two-factor-authentication is not in use, it is (in theory) possible to brute-force user passwords and thus gain access to their account. One, relatively easy, way to mitigate this, is setting up fail2ban which blocks ipadresses after too many failed login attempts. However: Care should be taken when using this behind multiple reverse-proxies (such as cloudflare). See: Fail2Ban Setup

Hiding under a subdir

Traditionally, a Bitwarden instance resides at the root of a subdomain (i.e., bitwarden.example.com, and not bitwarden.example.com/some/path). The upstream Bitwarden server currently only supports subdomain roots, while vaultwarden adds support for alternate base directories. For some users, this is useful simply because they only have access to one subdomain and want to run multiple services under different directories. In such cases, they typically choose something obvious like mysubdomain.example.com/bitwarden. However, you can also use this to provide an extra layer of protection by putting vaultwarden under something like mysubdomain.example.com/bitwarden/<mysecretstring>, where <mysecretstring> effectively acts as a password. Some may argue that this is security through obscurity, but it's actually defense in depth -- the secrecy of the subdir is just an extra layer of security, and not intended to be the primary means of security (which is still the strength of a user's master password).

For general discussion about subpath hosting for security refer to: https://github.com/debops/debops/issues/1233

If you'd like to make Caddy drop all connections besides for vaultwarden

mysubdomain.example.com {
	route {
		reverse_proxy /my-custom-path/* 10.0.0.150:8083 {
			header_up X-Real-IP {remote_host}
		}
		handle /* {
			abort
		}
	}
}