They will only be used in tests, but it doesn't play
nice with copy operations many tests use. The linter was
not happy. While the global clock needs mutexes for parallelism,
local Clocks only used it for Set/Add and didn't even use the
mutex for actual time functions.
The reflect.DeepCopy doesn't play nice with the new Lock and Clock
fields in sessions. And it added unneeded session deserialization
logic to every request.
* Add sensible logging flag to default setup for logger
* Add Redis lock
* Fix default value flag for sensitive logging
* Split RefreshSessionIfNeeded in two methods and use Redis lock
* Small adjustments to doc and code
* Remove sensible logging
* Fix method names in ticket.go
* Revert "Fix method names in ticket.go"
This reverts commit 408ba1a1a5c55a3cad507a0be8634af1977769cb.
* Fix methods name in ticket.go
* Remove block in Redis client get
* Increase lock time to 1 second
* Perform retries, if session store is locked
* Reverse if condition, because it should return if session does not have to be refreshed
* Update go.sum
* Update MockStore
* Return error if loading session fails
* Fix and update tests
* Change validSession to session in docs and strings
* Change validSession to session in docs and strings
* Fix docs
* Fix wrong field name
* Fix linting
* Fix imports for linting
* Revert changes except from locking functionality
* Add lock feature on session state
* Update from master
* Remove errors package, because it is not used
* Only pass context instead of request to lock
* Use lock key
* By default use NoOpLock
* Remove debug output
* Update ticket_test.go
* Map internal error to sessions error
* Add ErrLockNotObtained
* Enable lock peek for all redis clients
* Use lock key prefix consistent
* Fix imports
* Use exists method for peek lock
* Fix imports
* Fix imports
* Fix imports
* Remove own Dockerfile
* Fix imports
* Fix tests for ticket and session store
* Fix session store test
* Update pkg/apis/sessions/interfaces.go
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
* Do not wrap lock method
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
* Use errors package for lock constants
* Use better naming for initLock function
* Add comments
* Add session store lock test
* Fix tests
* Fix tests
* Fix tests
* Fix tests
* Add cookies after saving session
* Add mock lock
* Fix imports for mock_lock.go
* Store mock lock for key
* Apply elapsed time on mock lock
* Check if lock is initially applied
* Reuse existing lock
* Test all lock methods
* Update CHANGELOG.md
* Use redis client methods in redis.lock for release an refresh
* Use lock key suffix instead of prefix for lock key
* Add comments for Lock interface
* Update comment for Lock interface
* Update CHANGELOG.md
* Change LockSuffix to const
* Check lock on already loaded session
* Use global var for loadedSession in lock tests
* Use lock instance for refreshing and releasing of lock
* Update possible error type for Refresh
Co-authored-by: Joel Speed <Joel.speed@hotmail.co.uk>
* Set and verify a nonce with OIDC
* Create a CSRF object to manage nonces & cookies
* Add missing generic cookie unit tests
* Add config flag to control OIDC SkipNonce
* Send hashed nonces in authentication requests
* Encrypt the CSRF cookie
* Add clarity to naming & add more helper methods
* Make CSRF an interface and keep underlying nonces private
* Add ReverseProxy scope to cookie tests
* Align to new 1.16 SameSite cookie default
* Perform SecretBytes conversion on CSRF cookie crypto
* Make state encoding signatures consistent
* Mock time in CSRF struct via Clock
* Improve InsecureSkipNonce docstring
* Centralize Ticket management of persistent stores
persistence package with Manager & Ticket will handle
all the details about keys, secrets, ticket into cookies,
etc. Persistent stores just need to pass Save, Load &
Clear function handles to the persistent manager now.
* Shift to persistence.Manager wrapping a persistence.Store
* Break up the Redis client builder logic
* Move error messages to Store from Manager
* Convert ticket to private for Manager use only
* Add persistence Manager & ticket tests
* Make a custom MockStore that handles time FastForwards
* Encode sessions with MsgPack + LZ4
Assumes ciphers are now mandatory per #414. Cookie & Redis sessions
can fallback to V5 style JSON in error cases. TODO: session_state.go
unit tests & new unit tests for Legacy fallback scenarios.
* Only compress encoded sessions with Cookie Store
* Cleanup msgpack + lz4 error handling
* Change NewBase64Cipher to take in an existing Cipher
* Add msgpack & lz4 session state tests
* Add required options for oauthproxy tests
More aggressively assert.NoError on all
validation.Validate(opts) calls to enforce legal
options in all our tests.
Add additional NoError checks wherever error return
values were ignored.
* Remove support for uncompressed session state fields
* Improve error verbosity & add session state tests
* Ensure all marshalled sessions are valid
Invalid CFB decryptions can result in garbage data
that 1/100 times might cause message pack unmarshal
to not fail and instead return an empty session.
This adds more rigor to make sure legacy sessions
cause appropriate errors.
* Add tests for legacy V5 session decoding
Refactor common legacy JSON test cases to a
legacy helpers area under session store tests.
* Make ValidateSession a struct method & add CHANGELOG entry
* Improve SessionState error & comments verbosity
* Move legacy session test helpers to sessions pkg
Placing these helpers under the sessions pkg removed
all the circular import uses in housing it under the
session store area.
* Improve SignatureAuthenticator test helper formatting
* Make redis.legacyV5DecodeSession internal
* Make LegacyV5TestCase test table public for linter
This helper method is only applicable for Base64 wrapped
encryption since it operated on string -> string primarily.
It wouldn't be used for pure CFB/GCM ciphers. After a messagePack
session refactor, this method would further only be used for
legacy session compatibility - making its placement in cipher.go
not ideal.
Have it take in a cipher init function as an argument.
Remove the confusing `newCipher` method that matched legacy behavior
and returns a Base64Cipher(CFBCipher) -- instead explicitly ask for
that in the uses.
All Encrypt/Decrypt Cipher implementations will now take
and return []byte to set up usage in future binary compatible
encoding schemes to fix issues with bloat encrypting to strings
(which requires base64ing adding 33% size)
* 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>
