It isn't used in any providers and we have future plans
to remove the specialness of PreferredUsername and make it
an optional field in the session.
User, Email & Groups will eventually be the only first class
fields on the session that are always set.
`findClaimsFromIDToken` would always have a `nil` access token and not be
able to hit the userinfo endpoint in Bearer case. If access token is nil,
default to legacy `session.Email = claim.Subject` that all JWT bearers used
to have, even if a valid profileURL is present.
This reverts to functionality before #499 where an OIDC
provider could be used with `--skip-jwt-bearer-tokens` and
tokens without an email or profileURL would still be valid.
This logic mirrors `middleware.createSessionStateFromBearerToken`
which used to be the universal logic before #499.
* Changed how gitlab-group is parsed, from string to []string
See #637
* Point out that gitlab-group can be a list
See #637
* Reflect to the []string change on pkg/apis/options/options.go
See #637
* Move cfg option gitlab_group to gitlab_groups
See #637
* Renamed Group to Groups
See #637
* Reflect the change on gitlab.go as well
See #637
* Added #639
* Added the author of #639 to the CHANGELOG
* Add the gitlab_groups env change to CHANGELOG.md
See #639
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
When using the configured provider JWT Verifier, it makes
sense to use the provider `CreateSessionStateFromBearerToken`
method. For any extra JWT Issuers, they should use a generic
default verifier.
* Refactor the utils package to other areas
Move cookieSession functions to cookie session store
& align the double implementation of SecretBytes to be
united and housed under encryption
* Remove unused Provider SessionFromCookie/CookieForSession
These implementations aren't used, these are handled in the cookie store.
* Add changelog entry for session/utils refactor
* Add -user-id-claim to support other claims than email
Fix#431 - This is a minimal change to allow the user to configure which claim is
the source of the "user ID".
- Add the option `user-id-claim` (defaults to email)
- OIDC extracts this claim into session.Email (to be renamed later)
- providers: add `CreateSessionStateFromBearerToken` with a default impl taken from
`GetJwtSession` and overridden by oidc to respect `user-id-claim`
Once #466 is merged, I can continue to rename SessionState.Email to .UserID
and add HTTP headers with a corresponding name.
* Apply suggestions from code review
Co-Authored-By: Joel Speed <Joel.speed@hotmail.co.uk>
* Review feedback: Don't extract claims manually
Instead, parse them twice - it might be sligtly slower but less bug-prone as the code evolves.
* Fix indentation
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
* feature: switch Azure AD graph API to Microsoft Graph API
* Update CHANGELOG
* Expand Breaking Changes notice
* Update CHANGELOG.md
Co-Authored-By: Joel Speed <Joel.speed@hotmail.co.uk>
* fix: use constant http method
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
when type asserting fails here, err is reassigned with nil and the
default block of the switch prints out <nil> in the error message. This
makes debugging a configuration or access token issue difficult
The particular error this surfaces is:
Response: {
"error": "unauthorized_client",
"error_description": "Client is unauthorized to retrieve access tokens using this method, or client not authorized for any of the scopes requested."
}
Signed-off-by: Josh Bielick <jbielick@gmail.com>
* feature: Implement graceful shutdown
Propagate the request context to the Redis client.
It is possible to propagate a context cancel to Redis client if the connection is closed by the HTTP client.
The redis.Cmdable cannot use WithContext, so added the Client interface to handle redis.Client and redis.ClusterClient transparently.
Added handling of Unix signals to http server.
Upgrade go-redis/redis to v7.
* Update dependencies
- Upgrade golang/x/* and google-api-go
- Migrate fsnotify import from gopkg.in to github.com
- Replace bmizerany/assert with stretchr/testify/assert
* add doc for wrapper interface
* Update CHANGELOG.md
* fix: upgrade fsnotify to v1.4.9
* fix: remove unnessary logging
* fix: wait until all connections have been closed
* refactor: move chan to main for testing
* add assert to check if stop chan is empty
* add an idiomatic for sync.WaitGroup with timeout
* Upgrade base image tags and golangci-lint
* Upgrade golang and golangci-lint in travis-ci
* fix: tests, pointed out by golangci-lint and format files
* Upgrade dependencies
* update changelog
* fix: tests related to https://github.com/pusher/oauth2_proxy/pull/418
* Separate tests using go version build tags
* Update CHANGELOG
* Revert "Separate tests using go version build tags"
This reverts commit 9b7e65eb90cae954dc7b6316345d3207205d488a.
* fix test to support go1.14 and go1.13
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
* cleaned up source to make golangci-lint pass
* providers/azure_test.go: use build in POST constant
* options_test.go: do not export unnecessary variables
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
* Add support for preferred username.
* Add missing TOC entries.
* Add note about preferred_username support.
* Adjust tests.
* Check on not implemented error for GetPreferredUsername() call.
Co-authored-by: Felix Fontein <felix@fontein.de>
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
* OIDC Token Refresh works without id_tokens
Addresses https://github.com/pusher/oauth2_proxy/issues/318
Refactoring the OIDC provider so that the refresh process works when there are
no id_tokens present in the response. Added unit tests to the oidc_test.go to prove
the redeem and refresh still work.
The expiry time of the session is now taken from the outh token expiry and not
the id_token (preventing stale access_tokens in sessions).
* Refactoring the to use a KeySetStub in the oidc_test.go. This allows the
elimination of the slightly contrived function passing elements used
previously. (This change is being applied to address the bug #318)
* Changes as per the PR comments and preparing for 5.x release
* Fixup changelog
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
======================================================
changelog note
[#274](https://github.com/pusher/oauth2_proxy/pull/274) Add github api pagination support (@toshi-miura ,@apratina)
======================================================
I didn't edit CHANGELOG.md.
Since # 102 was taken over and the change difference of CHANGELOG.md was large
* Remove shadowing of predeclared identifier: new.
* strings.ReplaceAll instead of strings.Replace with -1.
* Change strings.ToLower comparison to strings.EqualFold.
* Rewrite if-else-if-else chain as a switch.
Add a new provider for Bitbucket,
can be configured from the options
specifying team and/or repository
that the user must be part/have access
to in order to grant login.
* Initial version of OIDC based GitLab provider
* Add support for email domain check to GitLab provider
* Add gitlab.com as default issuer for GitLab provider
* Update documentation for GitLab provider
* Update unit tests for new GitLab provider implementation
* Update CHANGELOG for GitLab provider
* Rename GitLab test access token as response to linter
* Check Google group membership with hasMember and get.
This PR is an enhancement built on
https://github.com/pusher/oauth2_proxy/pull/160. That PR reduces the
number of calls to the Google Admin API and simplifies the code by
using the hasMember method. It also supports checking membership in
nested groups.
However, the above message doesn't handle members who are not a part
of the domain. The hasMember API returns a 400 for that case. As a
fallback, when the API returns a 400, this change will try using the
`get` API which works as expected for members who aren't a part of the
domain. Supporting members who belong to the Google group but aren't
part of the domain is a requested feature from
https://github.com/pusher/oauth2_proxy/issues/95.
https://developers.google.com/admin-sdk/directory/v1/reference/members/get
Note that nested members who are not a part of the domain will not be
correctly detected with this change.
* Update CHANGELOG.
* Fix incorrect JSON and stop escaping strings.
* Add comments for each scenario.
* Current OIDC implementation asserts that user email check must come
from JWT token claims. OIDC specification also allows for source
of user email to be fetched from userinfo profile endpoint.
http://openid.net/specs/openid-connect-core-1_0.html#UserInfo
* First, attempt to retrieve email from JWT token claims. Then fall back to
requesting email from userinfo endpoint.
* Don't fallback to subject for email
https://github.com/bitly/oauth2_proxy/pull/481
Fix an unsupported protocol scheme error when validating tokens by
ensuring that the ValidateURL generates a non-empty string. The Azure
provider doesn't define any ValidateURL and therefore uses the default
value of `url.Parse("")` which is not `nil`.
The following log summary shows the issue:
2019/06/14 12:26:04 oauthproxy.go:799: 10.244.1.3:34112 ("10.244.1.1") refreshing 16h26m29s old session cookie for Session{email:jonas.fonseca@example.com user:jonas.fonseca token:true} (refresh after 1h0m0s)
2019/06/14 12:26:04 internal_util.go:60: GET ?access_token=eyJ0...
2019/06/14 12:26:04 internal_util.go:61: token validation request failed: Get ?access_token=eyJ0...: unsupported protocol scheme ""
2019/06/14 12:26:04 oauthproxy.go:822: 10.244.1.3:34112 ("10.244.1.1") removing session. error validating Session{email:jonas.fonseca@example.com user:jonas.fonseca token:true}
* Use encoding/json for SessionState serialization
In order to make it easier to extend in future.
* Store only email and user in cookie when cipher is unavailable
This improves safety and robustness, and also preserves the existing
behaviour.
* Add TestEncodeSessionState/TestDecodeSessionState
Use the test vectors with JSON encoding just introduced.
* Support session state encoding in older versions
* Add test cases for legacy session state strings
* Add check for wrong expiration time in session state strings
* Avoid exposing time.Time zero value when encoding session state string
* Update CHANGELOG.md
* first stab at login.gov provider
* fixing bugs now that I think I understand things better
* fixing up dependencies
* remove some debug stuff
* Fixing all dependencies to point at my fork
* forgot to hit save on the github rehome here
* adding options for setting keys and so on, use JWT workflow instead of PKCE
* forgot comma
* was too aggressive with search/replace
* need JWTKey to be byte array
* removed custom refresh stuff
* do our own custom jwt claim and store it in the normal session store
* golang json types are strange
* I have much to learn about golang
* fix time and signing key
* add http lib
* fixed claims up since we don't need custom claims
* add libs
* forgot ioutil
* forgot ioutil
* moved back to pusher location
* changed proxy github location back so that it builds externally, fixed up []byte stuff, removed client_secret if we are using login.gov
* update dependencies
* do JWTs properly
* finished oidc flow, fixed up tests to work better
* updated comments, added test that we set expiresOn properly
* got confused with header and post vs get
* clean up debug and test dir
* add login.gov to README, remove references to my repo
* forgot to remove un-needed code
* can use sample_key* instead of generating your own
* updated changelog
* apparently golint wants comments like this
* linter wants non-standard libs in a separate grouping
* Update options.go
Co-Authored-By: timothy-spencer <timothy.spencer@gsa.gov>
* Update options.go
Co-Authored-By: timothy-spencer <timothy.spencer@gsa.gov>
* remove sample_key, improve comments related to client-secret, fix changelog related to PR feedback
* github doesn't seem to do gofmt when merging. :-)
* update CODEOWNERS
* check the nonce
* validate the JWT fully
* forgot to add pubjwk-url to README
* unexport the struct
* fix up the err masking that travis found
* update nonce comment by request of @JoelSpeed
* argh. Thought I'd formatted the merge properly, but apparently not.
* fixed test to not fail if the query time was greater than zero
