This utility allows an easy way to add end-to-end cryptographic attestation to patches sent via mail. It does so by adapting the DKIM email signature standard to include cryptographic signatures via the X-Developer-Signature email header.
If your project workflow doesn't use patches sent via email, then you don't need this and should simply start signing your tags and commits.
DKIM is a widely adopted standard for domain-level attestation of email messages. It works by hashing the message body and certain individual headers, and then creating a cryptographic signature of the resulting hash. The receiving side obtains the public key of the sending domain from its DNS record and checks the signature and header/body hashes. If the signature verifies and the resulting hashes are identical, then there is a high degree of assurance that neither the body of the message nor any of the signed headers were modified in transit.
This utility uses the exact same DKIM standard to hash the headers and the body of the patch message, but uses a different set of fields and canonicalization routines:
- the d= field is not used (no domain signatures involved)
- the q= field is not used (key lookup is left to the client)
- the c= field is not used (see below for canonicalization)
- the i= field is optional, but MUST be the canonical email address of the sender, if not the same as the From: field
Patatt uses the "relaxed/simple" canonicalization as defined by the DKIM standard, but the message is first parsed by the "git-mailinfo" command in order to achieve the following:
- normalize any content-transfer-encoding modifications (convert back from base64/quoted-printable/etc into 8-bit)
- use any encountered in-body From: and Subject: headers to rewrite the outer message headers
- perform the subject-line normalization in order to strip content not considered by git-am when applying the patch (i.e. drop [PATCH .*] and other bracketed prefix content)
To achieve this, the message is passed through git-mailinfo with the following flags:
cat orig.msg | git mailinfo --encoding=utf-8 --no-scissors m p > i
Patatt then uses the data found in "i" to replace the From: and Subject: headers of the original message, and concatenates "m" and "p" back together to form the body of the message, which is then normalized using CRLF line endings and the DKIM "simple" body canonicalization (any trailing blank lines are removed).
Any other headers included in signing are modified using the "relaxed" header canonicalization routines as defined in the DKIM RFC.
In other words, the body and some of the headers are normalized and reconstituted using the "git-mailinfo" command, and then canonicalized using DKIM's relaxed/simple standard.
DKIM standard mostly relies on RSA signatures, though RFC 8463 extends it to support ED25519 keys as well. While it is possible to use any of the DKIM-defined algorithms, patatt only supports the following signing/hashing schemes:
- ed25519-sha256: exactly as defined in RFC8463
- openpgp-sha256: uses OpenPGP to create the signature
- openssh-sha256: uses OpenSSH signing capabilities
Note: Since GnuPG supports multiple signing key algorithms, openpgp-sha256 signatures can be done using EDDSA keys as well. However, since OpenPGP output includes additional headers, the "ed25519-sha256" and "openpgp-sha256" schemes are not interchangeable even when ed25519 keys are used in both cases.
Note: OpenSSH signature support was added in OpenSSH 8.0 and requires ssh-keygen that supports the -Y flag.
In the future, patatt may add support for more algorithms, especially if that allows incorporating more hardware crypto offload devices (such as TPM).
Patatt adds a separate X-Developer-Key: header with public key
information. It is merely informational and ISN'T and SHOULDN'T be used
for performing any kind of message validation (for obvious reasons). It
is included to make it easier for maintainers to obtain the
contributor's public key before performing whatever necessary
verification steps prior to its inclusion into their individual or
project-wide keyrings.
This also allows keeping a historical record of contributor keys via list archive services such as lore.kernel.org and others.
It is very easy to start signing your patches with patatt.
You can install from pip:
pip install --user patatt
Make sure your PATH includes $HOME/.local/bin.
Alternatively, you can clone this repository and symlink patatt.sh into your path:
cd bin ln -s ~/path/to/patatt/patatt.sh patatt
After this, you should be able to run patatt --help without
specifying the full path to the repository.
If you already have a PGP key, you can simply start using it to sign patches. Add the following to your ~/.gitconfig:
[patatt]
signingkey = openpgp:KEYID
The KEYID should be the 16-character identifier of your key, for example:
[patatt]
signingkey = openpgp:E63EDCA9329DD07E
If you have OpenSSH version 8.0+, then you can use your ssh keys for generating and verifying signatures. There are several upsides to using openssh as opposed to generic ed25519:
- you can passphrase-protect your ssh keys
- passphrase-protected keys will benefit from ssh-agent caching
- you can use hardware tokens and ed25519-sk keys for higher protection
- you are much more likely to remember to back up your ssh keys
To start using openssh signatures with patatt, add the following to your ~/.gitconfig:
[patatt]
signingkey = openssh:~/.ssh/my_key_id.pub
selector = my_key_id
Note, that the person verifying openssh signatures must also run the version of openssh that supports this functionality.
If you don't already have a PGP key, you can opt to generate and use a new ed25519 key instead (see below for some considerations on pros and cons of PGP vs ed25519 keys).
To generate a new keypair, run:
patatt genkey
You will see an output similar to the following:
Generating a new ed25519 keypair
Wrote: /home/user/.local/share/patatt/private/20210505.key
Wrote: /home/user/.local/share/patatt/public/20210505.pub
Wrote: /home/user/.local/share/patatt/public/ed25519/example.org/user/default
Add the following to your .git/config (or global ~/.gitconfig):
---
[patatt]
signingkey = ed25519:20210505
---
Next, communicate the contents of the following file to the
repository keyring maintainers for inclusion into the project:
/home/user/.local/share/patatt/public/20210505.pub
Please make sure to back up your new private key, located in
~/.local/share/patatt/private. It is short enough to simply
print/write out for storing offline.
Next, just do as instructions say. If the project for which you are contributing patches already uses patatt attestation, please work with the project maintainers to add your public key to the repository. If they aren't yet using patatt, just start signing your patches and hopefully the project will start keeping its own keyring in the future.
To test if it's working:
$ git format-patch -1 --stdout | patatt sign > /tmp/test
If you didn't get an error message, then the process was successful. You
can review /tmp/test to see that X-Developer-Signature and
X-Developer-Key headers were successfully added.
You can now validate your own message:
$ patatt validate /tmp/test
If everything is working well, you can start automatically signing all outgoing patches sent via git-send-email. Inside the repo you want enabled for signing, run:
$ patatt install-hook
Or you can do it manually:
$ echo 'patatt sign --hook "${1}"' > "$(git rev-parse --git-dir)/hooks/sendemail-validate"
$ chmod a+x "$(git rev-parse --git-dir)/hooks/sendemail-validate"
If you don't already have a PGP key that is used in your project, you may wonder whether it makes sense to create a new PGP key, reuse your OpenSSH key, or start using standalone ed25519 keys.
Reasons to choose PGP:
- you can protect the PGP private key with a passphrase (gpg-agent will manage it for you so you only need to enter it once per session)
- you can move your PGP key to an OpenPGP-compliant smartcard to further protect your key from being leaked/stolen
- you can use PGP keys to sign git tags/commits, not just mailed patches
If you choose to create a new PGP key, you can use the following guide: https://github.com/lfit/itpol/blob/master/protecting-code-integrity.md
Reasons to choose OpenSSH keys:
- you can protect openssh keys with a passphrase and rely on ssh-agent passphrase caching
- you can use ssh keys with u2f hardware tokens for additional protection of your private key data
- very recent versions of git can also use ssh keys to sign tags and commits
Reasons to choose a standalone ed25519 key:
- much smaller signatures, especially compared to PGP RSA keys
- implements the DKIM ed25519 signing standard
- faster operation
If you choose ed25519 keys, you will need to make sure that PyNaCl is installed (pip install should have already taken care of it for you).
Patatt implements basic signature validation, but it's a tool aimed primarily at contributors. If you are processing mailed-in patches, then you should look into using b4, which aims at making the entire process easier. B4 properly recognizes X-Developer-Signature headers starting with version 0.7.0 and uses the patatt library as well.
That said, keyring management as discussed below applies both to patatt and b4, so you can read on for an overview.
The trickiest part of all decentralized PKI schemes is not the crypto itself, but public key distribution and management. PGP famously tried to solve this problem by relying on cross-key certification and keyservers, but the results were not encouraging.
On the other hand, within the context of git repositories, we already have a suitable mechanism for distributing developer public keys, which is the repository itself. Consider this:
- git is already decentralized and can be mirrored to multiple locations, avoiding any single points of failure
- all contents are already versioned and key additions/removals can be audited and "git blame'd"
- git commits themselves can be cryptographically signed, which allows a small subset of developers to act as "trusted introducers" to many other contributors (mimicking the "keysigning" process)
The idea of using git itself for keyring management was originally suggested by the did:git project, though we do not currently implement the proposed standard itself.
The keyring is structured as follows:
- dir: topdir (e.g. ".keys")
|
- dir: keytype (e.g. "ed25519" or "openpgp")
|
- dir: address-domainname (e.g. "example.org")
|
- dir: address-localpart (e.g. "developer")
|
- file: selector (e.g. "default")
The main reasoning behind this structure was to make it easy for multiple project maintainers to manage keys without causing any unnecessary git merge complications. Keeping all public keys in individual files helps achieve this goal.
For example, let's take the following signature:
From: Konstantin Ryabitsev <konstantin@linuxfoundation.org> X-Developer-Signature: v=1; a=ed25519-sha256; t=1620240207; l=2577; h=from:subject; bh=yqviDBgyf3/dQgHcBe3B7fTP39SuKnYInPBxnOiuGcA=; b=Xzd0287MvPE9NLX7xbQ6xnyrvqQOMK01mxHnrPmm1f6O7KKyogc8YH6IAlwIPdo+jk1CkdYYQsyZ sS0cJdX2B4uTmV9mxOe7hssjtjLcj5/NU9zAw6WJARybaNAKH8rv
The key would be found in the following subpath:
.keys/ed25519/linuxfoundation.org/konstantin/default
If i= and s= fields are specified in the signature, as below:
X-Developer-Signature: v=1; a=ed25519-sha256; t=1620244687; l=12645; i=mricon@kernel.org; s=20210505; h=from:subject; bh=KRCBcYiMdeoSX0l1XJ2YzP/uJhmym3Pi6CmbN9fs4aM=; b=sSY2vXzju7zU3KK4VQ5vFa5iPpDr3nrf221lnpq2+uuXmCODlAsgoqDmjKUBmbPtlY1Bcb2N0XZQ 0KX+OShCAAwB5U1dtFtRnB/mgVibMxwl68A7OivGIVYe491yll5q
Then the path would reflect those parameters:
.keys/ed25519/kernel.org/mricon/20210505
In the case of ed25519 keys, the contents of the file are just the base64-encoded public key itself. For openpgp keys, the format should be the ascii-armored public key export, for example obtained by using the following command:
gpg -a --export --export-options export-minimal keyid
For openssh keys, the key contents are a single line in the usual openssh pubkey format, e.g.:
ssh-ed25519 AAAAC3N... comment@or-hostname
It does not really make sense to require cryptographic attestation for patches submitted by occasional contributors. The only keys added to the keyring should be those of the core maintainers who have push access to the "canonical" repository location, plus the keys belonging to regular contributors with a long-term ongoing relationship with the project.
For smaller repositories with a handful of core maintainers, it makes sense to keep the keyring in the main branch, together with all other project files.
For large teams with thousands of regular contributors and teams of subsystem maintainers (e.g. the Linux kernel), it does not make sense to have a centrally managed keyring tracked in the main repository. Instead, each subsystem maintainer team should manage their own keyring in a separate ref of their own repository.
For example, to create a blank new ref called refs/meta/keyring:
git symbolic-ref HEAD refs/meta/keyring git reset --hard mkdir ed25519 openpgp
Individual public key files can then be added and committed following
the same structure as described above. Keeping the keyring outside the
regular development branch ensures that it doesn't interfere with
submitted pull requests or git-format-patch operations. Keeping the ref
under refs/meta/ will hide it from most GUI interfaces, but if that
is not the goal, then it can be stored in refs/heads just like any
other branch.
To commit and push the files after adding them, regular git operations should be used:
git commit -asS git push origin HEAD:refs/meta/keyring # Switch back to the development environment git checkout regular-branch
To make changes to an existing keyring ref, a similar workflow can be used:
git fetch origin refs/meta/keyring # Verify that the commit is signed git verify-commit FETCH_HEAD git checkout FETCH_HEAD # make any changes to the keys git commit -asS git push origin HEAD:refs/meta/keyring git checkout regular-branch
Alternatively, if key additions/updates are frequent enough, the remote ref can be checked out into its own workdir and set up for proper remote tracking.
To use the keyring with patatt or b4, just tell them which paths to
check, via the keyringsrc setting (can be specified multiple
times and will be checked in the listed order):
[patatt]
# Empty ref means "use currently checked out ref in this repo"
keyringsrc = ref:::.keys
# Use a dedicated ref in this repo called refs/meta/keyring
keyringsrc = ref::refs/meta/keyring:
# Use a ref in a different repo
keyringsrc = ref:~/path/to/another/repo:refs/heads/main:.keys
# Use a regular dir on disk
keyringsrc = ~/git/korg-pgpkeys/.keyring
For b4, use the same configuration under the [b4] section.
Any path on disk can be used for a keyring location, and some will always be checked just in case. The following locations are added by default:
ref:::.keys ref:::.local-keys ref::refs/meta/keyring: $XDG_DATA_HOME/patatt/public
The ":::" means "whatever ref is checked out in the current repo", and $XDG_DATA_HOME usually points at $HOME/.local/share.
Please send patches and support requests to tools@linux.kernel.org.
Submissions must be made under the terms of the Linux Foundation certificate of contribution and should include a Signed-off-by: line. Please read the DCO file for full legal definition of what that implies.
Why is this library even needed? Why not...
PGP-signing patches causes important problems for reviewers. If a patch is inline-signed, then this not only adds textual headers/footers, but adds additional escaping in the protected body, converting all '^-' sequences into '^- -', which corrupts patches.
MIME-signing is better, but has several other downsides:
- messages are now sent as multipart mime structures, which causes some tooling to no longer properly handle the patch content
- the signature attachments may be stripped/quarantined by email gateways that don't properly recognize OpenPGP mime signatures
- the From/Subject headers are rarely included into protected content, even though they are crucial parts of what ends up going into a git commit
These considerations have resulted in many projects specifically requesting that patches should NOT be sent PGP-signed.
Code review is a crucial step of the development process and patatt does not aim to replace it. However, there are several areas where the process can be abused by malicious parties in the absence of end-to-end cryptographic attestation:
- A maintainer who struggles with code review volume may delegate parts of their duties to a submaintainer. If that person submits aggregated patch series to the maintainer after performing that work, there must be a mechanism to ensure that none of the reviewed patches have been modified between when they were reviewed by the trusted submaintainer and when the upstream developer applies them to their tree. Up to now, the only mechanism to ensure this was via signed pull requests -- with patatt this is now also possible with regular patch series.
- It is important to ensure that what developer reviews is what actually ends up being applied to their git tree. Linux development process consists of collecting follow-up trailers (Tested-by, Reviewed-by, etc), so various tooling exists to aggregate these trailers and create the collated patch series containing all follow-up tags (see b4, patchwork, etc). Patatt signing provides a mechanism to ensure that what that developer reviewed and approved and what they applied to their tree is the exact same code and hasn't been maliciously modified in-between review and "git am" (e.g. by archival services such as lore.kernel.org, mail hosting providers, someone with access to the developer's inbox, etc).
- An attacker may attempt to impersonate a well-known developer by submitting malicious code, perhaps with the hope that it receives less scrutiny and is accepted without rigorous code review. Even if this attempt is unsuccessful (and it most likely would be), this may cause unnecessary reputation damage to the person being impersonated. Cryptographic signatures (and lack thereof) will help the developer quickly establish that the attack was performed without their involvement.
DKIM standard is great, but there are several places where it falls a bit short when it comes to patch attestation:
- The signing is done by the mail gateways that may or may not be properly checking that the "From:" header matches the identity of the authenticated user. For example, a service that allows free account registration may not check that alice@example.org sends outgoing email with "bob@example.org" in the "From:" field, which would allow Alice to impersonate Bob and have the messages arrive with a valid DKIM signature.
- DKIM is usually seen as merely a spam reduction mechanism, so there's usually little incentive for infrastructure administrators to be too strict about how they handle the private keys used for DKIM signing. Most likely, they are just stored on disk without a passphrase and accessible by the SMTP daemon.
- DKIM's "relaxed" canonicalization standard for message bodies replaces all multiple whitespace characters with a single space before the body hash is signed. This poses significant problems for patches where whitespace is syntactically significant (Python, Makefiles, etc). A "return True" with a different indent will pass DKIM signature check and may introduce a serious security vulnerability.
- DKIM doesn't prevent typosquatting attacks. For example, an attacker attempting to impersonate known.developer@companyname.com may send an email from known.developer@company-name.com or any other similar-looking address or domain, with valid DKIM signatures in every case.