👻
Ghostunnel is a simple TLS proxy with mutual authentication support for securing non-TLS backend applications.
Ghostunnel supports two modes, client mode and server mode. Ghostunnel in server mode runs in front of a backend server and accepts TLS-secured connections, which are then proxied to the (insecure) backend. A backend can be a TCP domain/port or a UNIX domain socket. Ghostunnel in client mode accepts (insecure) connections through a TCP or UNIX domain socket and proxies them to a TLS-secured service. In other words, ghostunnel is a replacement for stunnel.
Supported platforms: Ghostunnel is developed primarily for Linux on x86-64
platforms, although it should run on any UNIX system that exposes SO_REUSEPORT
,
including Darwin (macOS), FreeBSD, OpenBSD and NetBSD. Ghostunnel also supports
running on Windows, though with a reduced feature set. We recommend running on
x86-64 to benefit from constant-time implementations of cryptographic algorithms
that are not available on other platforms.
See ghostunnel --help
, ghostunnel server --help
and ghostunnel client --help
.
Access control: Ghostunnel enforces mutual authentication by requiring a valid client certificate for all connections. We also support access control via checks on the subject (or subject alternative names) of a client certificate. This is useful for restricting access to services that don't have native access control.
Certificate hotswapping: Ghostunnel can reload
certificates at runtime without dropping existing connections. Certificate
reloading can be triggered with a signal or on a regular time interval. This
allows short-lived certificates to be used with ghostunnel, new certificates
will get picked up transparently. And on platforms with SO_REUSEPORT
support,
restarts can be done with minimal downtime.
Monitoring and metrics: Ghostunnel has a built-in status feature that can be used to collect metrics and monitor a running instance. Metrics can be fed into Graphite (or other systems) to see number of open connections, rate of new connections, connection lifetimes, timeouts, and other info.
Emphasis on security: We have put some thought into making ghostunnel secure by default and prevent accidental misconfiguration. For example, we always negotiate TLS v1.2 and only use safe cipher suites. Ghostunnel also supports PKCS#11 which makes it possible to use Hardware Security Modules (HSMs) to protect private keys, and we have a bug bounty that pays rewards for security findings.
To get started and play around with the implementation, you will need to
generate some test certificates. If you want to bootstrap a full PKI, one
good way to get started is to use a package like
square/certstrap. If you only need
some test certificates for playing around with the tunnel, you can find
some pre-generated ones in the test-keys
directory (alongside instructions
on how to generate new ones with OpenSSL).
Ghostunnel is available through GitHub releases and through Docker Hub.
Binaries can be built from source as follows (cross-compile requires Docker and xgo):
# Compile for local architecture
make ghostunnel
# Cross-compile release binaries
make -f Makefile.dist dist
Note that ghostunnel requires Go 1.12 or later to build, and CGO is required for PKCS#11 support. See also CROSS-COMPILE for instructions on how to cross-compile a custom build with CGO enabled.
Ghostunnel has an extensive suite of integration tests. Our integration test suite requires Python 3.5 (or later) and gocovmerge to run. We use Go modules for managing vendored dependencies.
To run tests:
# Option 1: run unit & integration tests locally
make test
# Option 2: run unit & integration tests in a Docker container
GO_VERSION=1.15 make docker-test
# Open coverage information in browser
go tool cover -html coverage-merged.out
For more information on how to contribute, please see the CONTRIBUTING file.
By default, ghostunnel runs in the foreground and logs to stderr. You can set
--syslog
to log to syslog instead of stderr. If you want to run ghostunnel
in the background, we recommend using a service manager such as systemd or
runit, or use a wrapper such as daemonize or dumb-init.
Ghostunnel accepts certificates in multiple different file formats.
The --keystore
flag can take a PKCS#12 keystore or a combined PEM file with the
certificate chain and private key as input (format is auto-detected). The --cert
/
--key
flags can be used to load a certificate chain and key from separate PEM files
(instead of a combined one).
Ghostunnel also supports loading identities from the macOS keychain or the SPIFFE Workload API and having private keys backed by PKCS#11 modules, see the "Advanced Features" section below for more information.
This is an example for how to launch ghostunnel in server mode, listening for
incoming TLS connections on localhost:8443
and forwarding them to
localhost:8080
. Note that while we use TCP sockets on localhost
in this
example, both the listen and target flags can also accept paths to UNIX domain
sockets as their argument.
To set allowed clients, you must specify at least one of --allow-all
,
--allow-cn
, --allow-ou
, --allow-dns
or --allow-uri
. All checks are made
against the certificate of the client. Multiple flags are treated as a logical
disjunction (OR), meaning clients can connect as long as any of the flags
matches (see ACCESS-FLAGS for more information). In
this example, we assume that the CN of the client cert we want to accept
connections from is client
.
Start a backend server:
nc -l localhost 8080
Start a ghostunnel in server mode to proxy connections:
ghostunnel server \
--listen localhost:8443 \
--target localhost:8080 \
--keystore test-keys/server-keystore.p12 \
--cacert test-keys/cacert.pem \
--allow-cn client
Verify that clients can connect with their client certificate:
openssl s_client \
-connect localhost:8443 \
-cert test-keys/client-combined.pem \
-key test-keys/client-combined.pem \
-CAfile test-keys/cacert.pem
Now we have a TLS proxy running for our backend service. We terminate TLS in ghostunnel and forward the connections to the insecure backend.
This is an example for how to launch ghostunnel in client mode, listening on
localhost:8080
and proxying requests to a TLS server on localhost:8443
.
Start a backend TLS server:
openssl s_server \
-accept 8443 \
-cert test-keys/server-combined.pem \
-key test-keys/server-combined.pem \
-CAfile test-keys/cacert.pem
Start a ghostunnel with a client certificate to forward connections:
ghostunnel client \
--listen localhost:8080 \
--target localhost:8443 \
--keystore test-keys/client-combined.pem \
--cacert test-keys/cacert.pem
Verify that we can connect to 8080
:
nc -v localhost 8080
Now we have a TLS proxy running for our client. We take the insecure local connection, wrap them in TLS, and forward them to the secure backend.
We can combine the above two examples to get a full tunnel. Note that you can start the ghostunnels in either order.
Start netcat on port 8001
:
nc -l localhost 8001
Start the ghostunnel server:
ghostunnel server \
--listen localhost:8002 \
--target localhost:8001 \
--keystore test-keys/server-combined.pem \
--cacert test-keys/cacert.pem \
--allow-cn client
Start the ghostunnel client:
ghostunnel client \
--listen localhost:8003 \
--target localhost:8002 \
--keystore test-keys/client-keystore.p12 \
--cacert test-keys/cacert.pem
Verify that we can connect to 8003
:
nc -v localhost 8003
Now we have a full tunnel running. We take insecure client connections, forward them to the server side of the tunnel via TLS, and finally terminate and proxy the connection to the insecure backend.
Ghostunnel supports different types of access control flags in both client and server modes. All checks are made against the certificate of the client or server. Multiple flags are treated as a logical disjunction (OR), meaning clients can connect as long as any of the flags matches. Ghostunnel is compatible with SPIFFE X.509 SVIDs.
See ACCESS-FLAGS for details.
You can silence specific types of log messages using the --quiet=...
flag,
such as --quiet=conns
or --quiet=handshake-errs
. You can pass this flag
repeatedly if you want to silence multiple different kinds of log messages.
Supported values are:
all
: silences all log messagesconns
: silences log messages about new and closed connections.conn-errs
: silences log messages about connection errors encountered (post handshake).handshake-errs
: silences log messages about failed handshakes.
In particular we recommend setting --quiet=handshake-errs
if you are
running TCP health checks in Kubernetes on the listening port, and you
want to avoid seeing error messages from aborted connections on each health
check.
To trigger a reload, simply send SIGUSR1
to the process or set a time-based
reloading interval with the --timed-reload
flag. This will cause ghostunnel
to reload the certificate and private key from the files on disk. Once
successful, the reloaded certificate will be used for new connections going
forward.
Additionally, ghostunnel uses SO_REUSEPORT
to bind the listening socket on
platforms where it is supported (Linux, Apple macOS, FreeBSD, NetBSD, OpenBSD
and DragonflyBSD). This means a new ghostunnel can be started on the same
host/port before the old one is terminated, to minimize dropped connections (or
avoid them entirely depending on how the OS implements the SO_REUSEPORT
feature).
Note that if you are using an HSM/PKCS#11 module, only the certificate will be reloaded. It is assumed that the private key in the HSM remains the same. This means the updated/reissued certificate much match the private key that was loaded from the HSM previously, everything else works the same.
Ghostunnel has a notion of "status port", a TCP port (or UNIX socket) that can
be used to expose status and metrics information over HTTPS. The status port
feature can be controlled via the --status
flag. Profiling endpoints on the
status port can be enabled with --enable-pprof
.
See METRICS for details.
Ghostunnel has support for loading private keys from PKCS#11 modules, which should work with any hardware security module that exposes a PKCS#11 interface.
See HSM-PKCS11 for details.
Ghostunnel has support for maintaining up-to-date, frequently rotated identities and trusted CA certificates from the SPIFFE Workload API.
See SPIFFE-WORKLOAD-API for details.
Ghostunnel supports socket activation via both systemd (on Linux) and launchd
(on macOS). Socket activation is support for the --listen
and --status
flags, and can be used by passing an address of the form systemd:<name>
or
launchd:<name>
, where <name>
should be the name of the socket as defined in
your systemd/launchd configuration.
See SOCKET-ACTIVATION for examples.
Ghostunnel in server mode supports signalling of transport connection information
to the backend using the PROXY protocol
(v2), just pass the --proxy-protocol
flag on startup. Note that the backend must
also support the PROXY protocol and must be configured to use it when setting
this option.
If ghostunnel has been compiled with build tag certstore
(off by default,
requires macOS 10.12+) a new flag will be available that allows for loading
certificates from the macOS keychain. This is useful if you have identities
stored in your local keychain that you want to use with ghostunnel, e.g. if you
want your private key(s) to be backed by the SEP on newer Touch ID MacBooks.
Certificates from the keychain can be loaded by selecting them based on the
Common Name (CN) of the subject.
For example, if you have an identity with CN 'example' in your login keychain:
ghostunnel client \
--keychain-identity example \
--listen unix:/path/to/unix/socket \
--target example.com:443 \
--cacert test-keys/cacert.pem
The command above launches a ghostunnel instance that uses the certificate and private key with Common Name 'example' from your login keychain to proxy plaintext connections from a given UNIX socket to example.com:443.