Crates.io | sequoia-chameleon-gnupg |
lib.rs | sequoia-chameleon-gnupg |
version | 0.11.2 |
source | src |
created_at | 2022-12-19 11:52:27.281177 |
updated_at | 2024-09-17 07:36:00.11895 |
description | Sequoia's reimplementation of the GnuPG interface |
homepage | https://sequoia-pgp.org/ |
repository | https://gitlab.com/sequoia-pgp/sequoia-chameleon-gnupg |
max_upload_size | |
id | 741143 |
size | 6,611,051 |
This is a re-implementation and drop-in replacement of gpg
and
gpgv
using the Sequoia OpenPGP implementation.
If you are a power user, you can try out the Chameleon today to see if it covers your use cases. The Chameleon does not directly modify any of GnuPG's data structures, so it is safe to try it out with your existing GnuPG installation and keys.
There are two ways the Chameleon will change your $GNUPGHOME
:
It will create an openpgp-cert-d overlay in
$GNUPGHOME/pubring.cert.d
(GnuPG will ignore this), unless you
are using the default $GNUPGHOME
, in which case the default
location for the openpgp-cert-d is used. This provides seamless
integration with other tools using the common certificate store,
such as Sequoia's native sq frontend.
If you create or import secret keys, the Chameleon will interact
with gpg-agent
the same way GnuPG would, and gpg-agent
will in
turn modify $GNUPGHOME
.
The Chameleon is compatible with the state of and gpg-agent
from
GnuPG 2.2.x and 2.4.x. If it discovers state of GnuPG < 2.1.x
(i.e. with secring.gpg
), it will convert it to 2.2.x state the same
way GnuPG would (i.e. import the secret keys into the gpg-agent
and
add a flag file).
To use the Chameleon, you need to install it, and then make sure that it is invoked either directly by you or indirectly by programs instead of GnuPG. There are three ways to do that.
If you only interact with GnuPG via the command line, then invoking
gpg-sq
instead of gpg
is enough to use the Chameleon instead of
g10code's GnuPG.
To globally replace g10code's gpg
with the Chameleon, install it as
/bin/gpg
(or /usr/bin/gpg
, or wherever gpg
is installed on your
system; hint: try type gpg
in your shell) replacing g10code's gpg
,
for example using dpkg-divert
or a similar mechanism. This is more
convenient and robust than invoking it using gpg-sq
, but has the
downside of being a system-wide all-or-nothing switch.
If the Chameleon can be found under the name gpg
in your $PATH
before g10code's gpg
is found, it takes precedence. This has the
advantage of being more convenient and robust than invoking it using
gpg-sq
, while allowing a more fine-grained replacement method (for
example per-shell if the $PATH
is manipulated on a per-shell basis,
or per-user if that is done in the shell's startup files).
If you go down this route, you also need to make sure that gpgconf
points to the Chameleon, because many programs invoke gpgconf
to
find the location of gpg
, notably those that use GPGME. To that
end, we have a shim that can be used from the build directory (if you
want to install the Chameleon, or your cargo target directory is
different, you need to adapt it accordingly):
$ export PATH=$(pwd)/shim-release:$PATH
$ gpg --version | head -n1
gpg (GnuPG-compatible Sequoia Chameleon) 2.2.40
$ gpgconf | head -n1
gpg:OpenPGP:.../sequoia-chameleon-gnupg/shim-release/gpg
You can at any point switch back to using g10code's gpg by undoing the replacement method discussed in the previous section.
However, a consequence of not modifying GnuPG's state but using an overlay is that changes made using the Chameleon will not be picked up by GnuPG. Therefore, you need to sync all the changes you made using the Chameleon back to g10code's GnuPG.
For example, if you import a certificate using the Chameleon, it will
only be inserted into the overlay, and g10code's GnuPG will not see
it. This is also true if you generate a key using the Chameleon:
while the secret bits will be known to the gpg-agent
, the public
certificate will not. Similarly, if you modify the trust database,
the changes will only be written to our overlay.
To sync the changes to the certificate store and trust database back, export changes from the Chameleon and import them into g10code's GnuPG:
$ gpg-sq --export | gpg-g10code --import
$ gpg-sq --export-ownertrust | gpg-g10code --import-ownertrust
If you are using the Chameleon and GnuPG side-by-side, it is recommended to either do state changing actions using GnuPG, or sync the changes, either manually or periodically in a cron job.
First, you need to install Sequoia's build dependencies, as well as the SQLite3 library. Then build the Chameleon from a checkout of this repository:
$ git clone https://gitlab.com/sequoia-pgp/sequoia-chameleon-gnupg.git
$ cd sequoia-chameleon-gnupg
$ cargo build --release
[...]
The above creates the gpg-sq
and gpgv-sq
executables, the manual
pages, and shell completions. By default, the manual pages and shell
completions are put into the cargo
target directory, but the exact
location is unpredictable. To write the assets to a predictable
location, set the environment variable ASSET_OUT_DIR
to a suitable
location.
Alternatively, you can change the cryptographic library that is used.
Note that this will change the build dependencies. Currently,
crypto-openssl
and crypto-cng
are supported, which select OpenSSL
and Windows CNG, respectively. To select a different backend, disable
the default features and activate the corresponding feature:
$ git clone https://gitlab.com/sequoia-pgp/sequoia-chameleon-gnupg.git
$ cd sequoia-chameleon-gnupg
$ cargo build --release --no-default-features --features=crypto-openssl
[...]
To run the tests, you also need Sequoia's command-line frontend sq
.
Then, you can run the tests using cargo
:
$ cargo test
If you have a program that uses GnuPG, and you want to see whether it works with the Chameleon, a good way to do that is to run the test suite (or if there is none, just use the program as usual) with a debug build of the Chameleon, and enable the invocation log. The log will log every invocation of the Chameleon with all the arguments given, and whether an error occurred or not.
$ cargo build
[...]
$ export PATH=$(pwd)/shim-debug:$PATH
$ # WARNING: this only works with a debug build!
$ export SEQUOIA_GPG_CHAMELEON_LOG_INVOCATIONS=/tmp/invocation.log
$ # Run your test suite here. This is an example:
$ (gpg --version ; gpg --lsign-key) >/dev/null 2>&1
$ cat $SEQUOIA_GPG_CHAMELEON_LOG_INVOCATIONS
814360: "gpg" "--version"
814360: success
814359: "gpg" "--lsign-key"
814359: Command aLSignKey is not implemented.
gpgv-sq
is feature-complete. Please report any problems you
encounter when replacing gpgv
with gpgv-sq
. gpgv-sq
is
stateless, you can switch to it by invoking gpgv-sq
instead of
gpgv
, and switch between the implementations without further
considerations.
gpg-sq
is not feature-complete. It currently implements a growing
subset of commonly used commands and options. Most of the signature
creation and verification commands, the encryption and decryption
commands, the key listing commands, and some miscellaneous commands
are implemented.
Currently, we rely on GnuPG's gpg-agent
to handle secret key
operations. This is very convenient for users who are already using
GnuPG, but it means that you have to have gpg-agent
installed.
Further, we don't have our own implementation of gpgconf
, and we
rely on it to start the gpg-agent
and downstream users, such as
GPGME, also require gpgconf
to function.
The Chameleon implements a subset of the trust models implemented by GnuPG, and some specific to Sequoia. The following table shows how the Chameleon and GnuPG interpret different values given to the "--trust-model" parameter.
Name | Chameleon | GnuPG |
---|---|---|
auto |
Auto | Auto |
sequoia |
Sequoia | ✗ |
sequoia+gnupg |
Sequoia+Gnupg | ✗ |
pgp |
Sequoia+GnuPG | GnuPG |
gnupg |
GnuPG | ✗ |
tofu |
✗ | Tofu |
tofu+pgp |
Sequoia+Gnupg | Tofu+GnuPG |
classic |
Sequoia+Gnupg | Classic |
direct |
✗ | Direct |
external |
✗ | External |
The trust models are described below.
The trust model is read from the trust database. If there is no trust
database, defaults to pgp
.
This is how Sequoia manages the trust roots. Under Sequoia's trust
model, only the trust-root
from the certificate directory is used as
trust root. Trust is managed using the sq pki
commands.
This model combines well with GnuPG's native trust model. This is
implemented in the sequoia+gnupg
trust model, which is the default.
However, if you are comfortable with using sq pki
to manage trust,
you can explicitly select the sequoia
trust model to deactivate the
gnupg
trust model, which is a little bit cheaper, and is easier to
reason about.
Combines the Sequoia trust model and the GnuPG trust model.
GnuPG uses the owner trust databases to select trust roots. In addition to ultimately trusted certs, also fully and marginally trusted certs are used as trust roots if they can be authenticated. This is the trust model used in PGP6.
Currently, we don't honor the constraint on the lengths of
certification chains (parameter --max-cert-depth
), but this will be
implemented at some point.
Trust-On-First-Use provides asymptotic trust, i.e. as the time goes on, and the observed binding between user IDs and certificates doesn't change, trust into the binding increases.
Combines the Tofu trust model with the GnuPG trust model.
Like the GnuPG trust model, but disregards trust delegations via trust signatures. This is the trust model of PGP2.
Uses owner trust values to validate authenticity of user IDs on certificates.
Uses the trust database, but refrains from modifying it. This is left to an external program.
Some features of g10code's GnuPG are deliberately not implemented to improve security, or to reduce complexity in the reimplementation.
This is a non-exhaustive list. If you note a divergence that is not listed here, please open an issue, so that we can either align with GnuPG's behavior, or add the divergence to this list.
Conversely, if you take issue with a divergence listed here, and have good arguments not mentioned here that support aligning with GnuPG on that aspect, please also open an issue.
We reject more weak algorithms than GnuPG by default, notably SHA-1 (see also this section).
The environment variable SEQUOIA_CRYPTO_POLICY
can point to a
configuration file (see crypto policy format) that changes which
cryptographic algorithms are acceptable. By default,
/etc/crypto-policies/back-ends/sequoia.config
is read, which on
Fedora contains a reasonable policy set by the distribution.
Short key IDs, aka 32-bit key IDs, are not supported. Creating keys with colliding short key IDs is trivial, so using them to refer to keys is insecure. See https://evil32.com .
When a short key ID is used in an command line argument or configuration file, an error is printed and the operation fails.
We don't use GnuPG's dirmngr
to connect to network services to
discover certificates. Instead, we read its configuration file and do
the requests ourselves. This is more robust and offers more features,
for example, if you have more than one keyserver
directive in
dirmngr.conf
, we will query all keyservers simultaneously.
We don't use GnuPG's keyboxd
to access keybox databases. Instead,
we read the certificates directly from the keybox database if we
discover one.
This flag is insecure and may overwrite arbitrary files with attacker-controlled content on decryption. We reject the flag by making the invocation fail. Do not use this flag.
GnuPG translates all messages and error messages, and even some
command-line arguments (LANGUAGE=de gpg --compression-algo unkomprimiert
...). We may translate strings in the future, but will
never do locale-dependent argument parsing.
Algorithms may be only be specified using their OpenPGP algorithm IDs and the symbolic identifiers as used by GnuPG. If you absolutely need this, please open an issue.
The Chameleon has a number of non-functional advantages relative to GnuPG.
The Chameleon includes a component called Parcimonie (after the venerable Parcimonie) that will keep your certificates up-to-date, trying to do so in a privacy preserving fashion.
It will periodically use any enabled auto-key-locate methods to search for updates in the local certificate store using randomized delays trying to de-correlate them. It will use Tor if available.
To enable the Parcimonie component, run gpg-sq --x-sequoia-parcimonie
, either manually or using a service manager (a
systemd unit file with the name gpg-sq-parcimonie.service
is
included in this repository). Alternatively, you can use the
x-sequoia-autostart-parcimonie
option in your configuration file to
start it on-demand if gpg-sq
is invoked.
Sequoia implements nearly all of the OpenPGP RFC4880. The missing bits are either obsolete or insecure. Furthermore, we engage with the IETF OpenPGP working group, and wrote an extensive OpenPGP Interoperability Test Suite to improve interoperability between various implementations. In short, if you use Sequoia to encrypt your data, you can be sure that you can decrypt it with any other OpenPGP implementation.
The Chameleon uses a very fast implementation of the Web-of-Trust,
and we calculate the trust on the fly without relying on a cache like
GnuPG. That means that gpg --check-trustdb
is a no-operation for
the Chameleon, whereas GnuPG is known to take a long time.
SHA-1 is broken. Unfortunately, SHA-1 is still widely used. To deal with this Sequoia implements a number of countermeasures:
Sequoia uses SHA1-CD, a variant of SHA-1 that detects and mitigates collision attacks. This protection is also used by GitHub, among others.
Sequoia rejects all signatures using SHA-1 by default.
On the other hand, GnuPG accepts SHA-1 everywhere without any additional protections.
Sequoia includes a salt in signatures and self-signatures to defend against collision attacks, among others. OpenSSH does the same thing. Should the collision resistance of another hash be broken, this will frustrate attackers trying to perform a Shambles-style attack.
Thanks to Rust's safer concurrency paradigms, it is less dangerous and less complicated for the Chameleon to use threads than implementations written in other languages. The Chameleon uses this, for instance, to parse keyrings faster.
There are two ways we test the Chameleon: we run the Chameleon and GnuPG side-by-side and compare the results, and we use test suites of programs that use GnuPG.
We run experiments that invoke the Chameleon and record human-readable
and machine-readable output and side-effects, and compare that to what
GnuPG emits. These tests are run when you invoke cargo test
, and
hence need GnuPG to be installed.
We use test suites of programs that directly or indirectly use GnuPG to verify that we support the required functionality.
A reoccurring problem when running these test suites is that they may include cryptographic artifacts such as OpenPGP certificates, keys, signatures, and messages. Those are rarely updated, and hence are stuck in time using old packet formats or insecure algorithms.
If you are maintaining a software package that includes cryptographic artifacts in the test suite, please help by regularly updating the artifacts. Further, try to reduce the amount of checked-in artifacts in the first place. Where possible, try to generate the required artifacts in the test.
This project was made possible and was funded through NGI Assure, a fund established by NLnet with financial support from the European Commission's Next Generation Internet program. Learn more at the NLnet project page.
Sequoia GnuPG Chameleon is free software: you can redistribute it
and/or modify it under the terms of the GNU General Public License as
published by the Free Software Foundation, either version 3 of the
License, or (at your option) any later version. See
LICENSE.txt
.
Sequoia GnuPG Chameleon is a derived work of GnuPG. It is not a
clean-room reimplementation, and it does include parts of GnuPG either
literally, or transcribed to Rust. Therefore, parties who claim
copyright to GnuPG also have a claim on parts of the Chameleon. See
AUTHORS.GnuPG
for a list.