* refactor(suites): replace selenium with go-rod
This change replaces [tebeka/selenium](https://github.com/tebeka/selenium) with [go-rod](https://github.com/go-rod/rod).
We no longer have a chromedriver/external driver dependency to utilise Selenium as we instead utilise the Chrome Dev Protocol to communicate with the browser.
Rod [documents](https://go-rod.github.io/#/why-rod) benefits of choosing the library as opposed to the available alternatives.
* feat(web): replace cra with vite
* fix: add istanbul
* fix: add jest
* fix: inject env vars
* fix: replicate cra output directories
* fix: post-frontend build for go templating
* fix: dynamic publicpath
* fix(web): import resolution with aliases for .module.css files
* refactor(server): baseurl var
* refactor(web): drop babel-jest for esbuild-jest
* refactor(web): add inline sourcemap for coverage bundle
* build(deps): update web deps
* build(deps): downgrade vite-plugin-istanbul to 2.2.0
98bf77dbaa is a breaking change that means production mode builds can no longer be instrumented.
* refactor(web): match frontend name and version
* refactor(web): drop cra readme
* ci(codecov): utilise new codecov uploader for coverage
The codecov bash uploader is being [deprecated](https://docs.codecov.com/docs/about-the-codecov-bash-uploader).
This utilises the new uploader which is recommended.
* ci(codecov): adjust file search path and name uploads
* fix(suites): coverage paths for codecov
* refactor: replace sqlite library with the original cgo variant
This change reverts our SQLite library back to the original for compatibility and performance reasons now that we always package with CGO.
* fix: cgo and build flags
* fix: gcc requirement
* fix: gcc in dev workflow too
* Check for pnpm in authelia-scripts
* Improve husky hooks to check for required apps
* Use pnpm in coverage dockerfile
* Use pnpm in dev workflow
* Stop buildx log truncation
* Ignore pnpm lockfile in yamllint
* Update versions required for docker and docker-compose in contributing docs
This adds method to validate the system clock is synchronized on startup. Configuration allows adjusting the server address, enabled state, desync limit, and if the error is fatal.
Co-authored-by: James Elliott <james-d-elliott@users.noreply.github.com>
This change means we only check the filters for the existence of placeholders that cannot be replaced at startup. We then utilized cached results of that lookup for subsequent replacements.
This commit replaces github.com/spf13/viper with github.com/knadh/koanf. Koanf is very similar library to viper, with less dependencies and several quality of life differences. This also allows most config options to be defined by ENV. Lastly it also enables the use of split configuration files which can be configured by setting the --config flag multiple times.
Co-authored-by: Amir Zarrinkafsh <nightah@me.com>
This change adjusts several global options moving them into the server block. It additionally notes other breaking changes in the configuration.
BREAKING CHANGE: Several configuration options have been changed and moved into other sections. Migration instructions are documented here: https://authelia.com/docs/configuration/migration.html#4.30.0
This adds a performance change to the default Active Directory users filter. Basically as per TechNet the (sAMAccountType=805306368) filter is the same as (&(objectCategory=person)(objectClass=user)) except the performance is better.
* This gives admins more control over their OIDC installation exposing options that had defaults before. Things like lifespans for authorize codes, access tokens, id tokens, refresh tokens, a option to enable the debug client messages, minimum parameter entropy. It also allows admins to configure the response modes.
* Additionally this records specific values about a users session indicating when they performed a specific authz factor so this is represented in the token accurately.
* Lastly we also implemented a OIDC key manager which calculates the kid for jwk's using the SHA1 digest instead of being static, or more specifically the first 7 chars. As per https://datatracker.ietf.org/doc/html/draft-ietf-jose-json-web-key#section-8.1.1 the kid should not exceed 8 chars. While it's allowed to exceed 8 chars, it must only be done so with a compelling reason, which we do not have.
#2101 introduced a minor regression when using the authelia scripts suite for developing.
The following issues occurred:
```
[00] # runtime/cgo
[00] cgo: exec gcc: exec: "gcc": executable file not found in $PATH
```
Adding the CGO_ENABLED=0 before the dlv build command in the run-backend-dev.sh fixed the issue.
* refactor: drop cgo requirement for sqlite
Replace github.com/mattn/go-sqlite3 with modernc.org/sqlite which drops our CGO requirement.
* refactor: newline for consistency with dockerfiles
This commit adjusts the build flags to include version information in the LDFLAGS using the -X options. Additionally this makes the information recorded at build time more comprehensive. All build information can now be obtained via the `authelia build` command, and the `authelia version` command is now `authelia --version`. Lastly this adjusts the Dockerfile to utilize docker cache more effectively.
* refactor: logging config key to log
This refactors the recent pre-release change adding log options to their own configuration section in favor of a log section (from logging).
* docs: add step to getting started to get the latest tagged commit
This is so we avoid issues with changes on master having differences that don't work on the latest docker tag.
* test: adjust tests
* docs: adjust doc strings
This is so levels like warn and error can be used to exclude info or warn messages. Additionally there is a reasonable refactoring of logging moving the log config options to the logging key because there are a significant number of log options now. This also decouples the expvars and pprof handlers from the log level, and they are now configured by server.enable_expvars and server.enable_pprof at any logging level.
OpenID connect has become a standard when it comes to authentication and
in order to fix a security concern around forwarding authentication and authorization information
it has been decided to add support for it.
This feature is in beta version and only enabled when there is a configuration for it.
Before enabling it in production, please consider that it's in beta with potential bugs and that there
are several production critical features still missing such as all OIDC related data is stored in
configuration or memory. This means you are potentially going to experience issues with HA
deployments, or when restarting a single instance specifically related to OIDC.
We are still working on adding the remaining set of features before making it GA as soon as possible.
Related to #189
Co-authored-by: Clement Michaud <clement.michaud34@gmail.com>
This change implements yamllint and adjusts all yaml files to abide by our linting setup. This excludes config.template.yml as this will be done in an alternate commit.
This PR changes the suites tests so if go test -short is used, they are skipped per go standards and a message is displayed. Additionally removed some redundant types from suite_high_availability_test.go and adjusted a warning about a nil req var.
Returns a 404 instead of 405 on bad API endpoints. The original issue was resolved in 3487fd392e however this resolves another issue that's related. Additionally this ensures the behavior is tested.
Co-authored-by: Clément Michaud <clement.michaud34@gmail.com>
Fixes#1520Closes#1534
* feat: go:embed static assets
Go 1.16 introduced the ability to embed files within a generated binary directly with the go tool chain. This simplifies our dependencies and the significantly improves the development workflow for future developers.
Key points to note:
Due to the inability to embed files that do not reside within the local package we need to duplicate our `config.template.yml` within `internal/configuration`.
To avoid issues with the development workflow empty mock files have been included within `internal/server/public_html`. These are substituted with the respective generated files during the CI/CD and build workflows.
* fix(suites): increase ldap suite test timeout
* fix(server): fix swagger asset CSP
This PR achieves the following goals:
* Utilise upstream version of kind instead of a patched version which allows binding to networks other than the default "kind"
* Utilises the registry cache which is setup one level above the kind cluster
The former point was required to successfully run our integration tests in a Kubernetes environment, however this is now possible without running a patched version of kind.
The second point is because DockerHub has introduced rate limiting for container downloads. If there are a large number of CI jobs nodes may occasionally be rejected due to the Kubernetes suite not pulling down from the registry cache.
renovate[bot]
Amir Zarrinkafsh
yossbg
James Elliott
Clément Michaud
Philipp Staiger
