Adds a workflow that compares ownerless files between the PR base and
head. Fails if the PR introduces files not covered by any team in
code-owners.json. Uses @actions/glob for pattern matching, consistent
with the existing code-owner-approval workflow.
- Add local wireguard go import
- Activate DAITA and add `wgActivateDaita` and `wgReceiveEvent` FFI
- Implement `start_daita` on Wireguard-go tunnel type
- Mention DAITA in `wireguard-go-rs` description
- Do not compile `wireguard-go-rs` on Windows
- Handle DAITA closed on `nil` event
- Handle daita action timeouts in libwg
- Remove noisy log lines
- Remove `maybenot_on_action` callback
- Remove unused link to `../build/lib` for `talpid-wireguard`
- Bump the `wireguard-go` submodule to a signed release tag in Mullvad's
`wireguard-go` fork.
- Update path to `libwg/go.sum` in verification script
Also:
- Use u64 instead of *mut void as log context
- Make Tunnel::set_config take a &mut self
- Use dyn Error instead of i32s for wg errors
Co-authored-by: Joakim Hulthe <joakim@hulthe.net>
The verification script running on github would complain that the first
commit of a PR changed all of the locked down files if this commit was
unsigned. The reason was that in our github enviornment we don't fetch
the entire git history but instead only the needed commits (so called
shallow repository). The entire history is therefore folded into the
first commit in the shallow repo. Since this history contains
modifications of locked down files and since this commit is not signed
the script will think this commit modified files its not allowed to.
This fix deepens the shallow repo by 1 after fetching.
It also clarifies a print in verification script to be more descriptive.
The github CI code broke due to a code injection vulnerability
being fixed in a quick-and-dirty way. The dirty solution is here
cleaned up and works better than previously.
Create a script which verifies that a set of "locked down" files are not
changed in commits that have not been signed.
Create a github workflow that runs the script in the CI.
The script accepts --whitelist <commit> and --import-gpg-keys arguments.
The default settings are supposed to work on the build server without
importing the gpg keys from the trusted_keys.pub file and running with a
hardcoded whitelist commit.
Make the CI workflow use these arguments as it is supposed to in
.github.
The public keys that can be imported are specified as files in the ci/keys/
directory.
The files that are locked down are specified in the .github workflow as
a single source of truth. This requires some complicated parsing in the
verification script as well as a dependency from the verification script
to the workflow YAML. These are not ideal design choices however the
alternative is to not have a single source of truth for the locked down
files as the github workflow can not depend on an external file.
The mullvad signing key is named to be first in the list in order to be
imported first.
The whitelisted commit is the latest master before this commit
Albin