* fix: handle undefined value in migration 66
* fix: the if-statement was typo
* feat: duplicate error behavior in fake storage service
* feat: fix all migrations that were setting undefined values
* feat: add test for disabled fingrint in migration 66
* fix: default single user state saving undefined value to state
* revert: awaiting floating promise
gonna fix this in a separate PR
* Revert "feat: fix all migrations that were setting undefined values"
This reverts commit 034713256c.
* feat: automatically convert save to remove
* Revert "fix: default single user state saving undefined value to state"
This reverts commit 6c36da6ba5.
* Require lifetime specification of user-scoped data
* Decouple tests for different classes
This coupling assumed constant interfaces with classes that isn't a guarantee and requires significant acrobatics to make types work, now that key definitions are not a consistent base.
* Fix types
* Do not update appId if it is not null
* Prefer linear transformations to side-effect-based changes
This leaves us open to repeat emits due to updates, but distinct until changed stops those.
Tracker improvements are due to passed in observables with replay causing immediate emits when `expectingEmission`s. This converts to a cold observable that only emits when the tracked observable does _after_ subscribing.
* Prefer while
* PR review
magical black-box key retrieval is slow, due to needing to repeatedly resolve (and potentially decrypt) keys. Doing the work up front is both more explicit and much faster.
It would be preferable to move OrganizationId to an opaque type in the models, but that has rippling effects all over the place and ultimately is stopped by vault filtering on strings rather than ids, but still calling the property `organizationId`, we need to fix that first.
* Match console method signatures in logService abstraction
* Add a few usages of improved signature
* Remove reality check test
* Improve electron logging
* Use account service to track accounts and active account
* Remove state service active account Observables.
* Add email verified to account service
* Do not store account info on logged out accounts
* Add account activity tracking to account service
* Use last account activity from account service
* migrate or replicate account service data
* Add `AccountActivityService` that handles storing account last active data
* Move active and next active user to account service
* Remove authenticated accounts from state object
* Fold account activity into account service
* Fix builds
* Fix desktop app switch
* Fix logging out non active user
* Expand helper to handle new authenticated accounts location
* Prefer view observable to tons of async pipes
* Fix `npm run test:types`
* Correct user activity sorting test
* Be more precise about log out messaging
* Fix dev compare errors
All stored values are serializable, the next step wasn't necessary and was erroring on some types that lack `toString`.
* If the account in unlocked on load of lock component, navigate away from lock screen
* Handle no users case for auth service statuses
* Specify account to switch to
* Filter active account out of inactive accounts
* Prefer constructor init
* Improve comparator
* Use helper methods internally
* Fixup component tests
* Clarify name
* Ensure accounts object has only valid userIds
* Capitalize const values
* Prefer descriptive, single-responsibility guards
* Update libs/common/src/state-migrations/migrate.ts
Co-authored-by: Justin Baur <19896123+justindbaur@users.noreply.github.com>
* Fix merge
* Add user Id validation
activity for undefined was being set, which was resulting in requests for the auth status of `"undefined"` (string) userId, due to key enumeration. These changes stop that at both locations, as well as account add for good measure.
---------
Co-authored-by: Justin Baur <19896123+justindbaur@users.noreply.github.com>
* Implement a lazy value class
This will be used as a source for composing key-protected storage from a single key source.
* Simplify local-backed-session-storage
The new implementation stores each value to a unique location, prefixed with `session_` to help indicate the purpose.
I've also removed the complexity around session keys, favoring passing in a pre-defined value that is determined lazily once for the service worker. This is more in line with how I expect a key-protected storage would work.
* Remove decrypted session flag
This has been nothing but an annoyance. If it's ever added back, it needs to have some way to determine if the session key matches the one it was written with
* Remove unnecessary string interpolation
* Remove sync Lazy
This is better done as a separate class.
* Handle async through type
* prefer two factory calls to incorrect value on races.
* Fix type
* Remove log
* Update libs/common/src/platform/misc/lazy.ts
Co-authored-by: Justin Baur <19896123+justindbaur@users.noreply.github.com>
---------
Co-authored-by: Justin Baur <19896123+justindbaur@users.noreply.github.com>
* Introduce browser large object storage location.
This location is encrypted and serialized to disk in order to allow for storage of uncountable things like vault items that take a significant amount of time to prepare, but are not guaranteed to fit within session storage.
however, limit the need to write to disk is a big benefit, so _most_ things are written to storage.session instead, where things specifically flagged as large will be moved to disk-backed memory
* Store derived values in large object store for browser
* Fix AbstractMemoryStorageService implementation
* remove active account unlocked from state service
* Remove status from account service `AccountInfo`
* Fixup lingering usages of status
Fixup missed factories
* Fixup account info usage
* fixup CLI build
* Fixup current account type
* Add helper for all auth statuses to auth service
* Fix tests
* Uncomment mistakenly commented code
* Rework logged out account exclusion tests
* Correct test description
* Avoid getters returning observables
* fixup type
* Use UserKeys in biometric state
* Remove global clear todo. Answer is never
* User UserKeys in crypto state
* Clear userkey on both lock and logout via User Key Definitions
* Use UserKeyDefinitions in environment service
* Rely on userKeyDefinition to clear org keys
* Rely on userKeyDefinition to clear provider keys
* Rely on userKeyDefinition to clear user keys
* Rely on userKeyDefinitions to clear user asym key pair
* Create tracker that can await until expected observables are received.
* Test dates are almost equal
* Remove unused class method
* Allow for updating active account in accout service fake
* Correct observable tracker behavior
Clarify documentation
* Transition config service to state provider
Updates the config fetching behavior to be lazy and ensure that any emitted value has been updated if older than a configurable value (statically compiled).
If desired, config fetching can be ensured fresh through an async.
* Update calls to config service in DI and bootstrapping
* Migrate account server configs
* Fix global config fetching
* Test migration rollback
* Adhere to implementation naming convention
* Adhere to abstract class naming convention
* Complete config abstraction rename
* Remove unnecessary cli config service
* Fix builds
* Validate observable does not complete
* Use token service to determine authed or unauthed config pull
* Remove superfluous factory config
* Name describe blocks after the thing they test
* Remove implementation documentation
Unfortunately the experience when linking to external documentation is quite poor. Instead of following the link and retrieving docs, you get a link that can be clicked to take you out of context to the docs. No link _does_ retrieve docs, but lacks indication in the implementation that documentation exists at all.
On the balance, removing the link is the better experience.
* Fix storybook
* Fix error on close due to context differences in background
Desktop background does not have active user information. Also, we want to delete _all_ prompt cancelled data, not just that for the active user. Storing this on global and manipulating observables to active achieves this without needing any user information in the background.
* Remove potentially orphaned data
* Throw nice error if prompt cancelled used without active user
* Register migration
* split prompt cancelled reset to user-specific and global
* Fix error on close due to context differences in background
Desktop background does not have active user information. Also, we want to delete _all_ prompt cancelled data, not just that for the active user. Storing this on global and manipulating observables to active achieves this without needing any user information in the background.
* Remove potentially orphaned data
* Throw nice error if prompt cancelled used without active user
* Register migration
* Fix nextMock arguments
* Add state for biometric prompts
* Use biometric state for prompts
* Migrate biometric prompt data
* wire up biometric state to logouts
* Add migrator to migrate list
* Remove usages of prompt automatically
Explicitly list non-nulled state as intentional
* `npm run prettier` 🤖
* Fix web lock component
* Move user key memory state to state providers
Note: state service observable change is because these updates are no longer internal to the class, but reporter directly to account service through crypto service on update of a user key
* remove decrypted user key state
Note, we're going to move the encrypted cryptoSymmetric key (and associated master key encrypted user keys) as part of the master key service creation. Crypto service will no longer be responsible for the encrypted forms of user key.
* Deprecate notices belong on abstraction
* Allow for single-direction status updates
This is necessary since we don't want to have to guarantee that the update to logged out occurs after the update to locked.
* Remove deprecated subject
It turns out the set for cryptoMasterKey was also unused 🎉
* Create state for biometric client key halves
* Move enc string util to central utils
* Provide biometric state through service
* Use biometric state to track client key half
* Create migration for client key half
* Ensure client key half is removed on logout
* Remove account data for client key half
* Remove unnecessary key definition likes
* Remove moved state from account
* Fix null-conditional operator failure
* Simplify migration
* Remove lame test
* Fix test type
* Add migrator
* Prefer userKey when legacy not needed
* Fix tests
* Add `disk-local` option for web
* Fix `web` DI
* Update libs/common/src/platform/state/state-definition.ts
Co-authored-by: Matt Gibson <mgibson@bitwarden.com>
* Rely On Default Implementation for Most of Cache Key
---------
Co-authored-by: Matt Gibson <mgibson@bitwarden.com>
* Expand state provider fakes
- default null initial value for fake states
- Easier mocking of key definitions through just the use of key names
- allows for not exporting KeyDefinition as long as the key doesn't collide
- mock of fake state provider to verify `get` calls
- `nextMock` for use of the fn mock matchers on emissions of `state$`
- `FakeAccountService` which allows for easy initialization and working with account switching
* Small bug fix for cache key collision on key definitions unique by only storage location
* Fix initial value for test
#7290 introduced these types, but during development we switched over to specifying dependencies in type parameters instead of an object. This change meant we no longer needed these `Type` or `ShapeToInstance` types, greatly simplifying the types related to derived state.
* Remove derived state from state classes
* Create provider for derived state
Derived state is automatically stored to memory storage, but can be derived from any observable.
* Fixup state provider method definitions
* Test `DefaultDerivedState`
* remove implementation notes
* Write docs for derived state
* fixup derived state provider types
* Implement buffered delayUntil operator
* Move state types to a common module
* Move mock ports to centra location
* Alias DerivedStateDependency type
* Add dependencies to browser
* Prefer internal rxjs operators for ref counting
* WIP
* Ensure complete on subjects
* Foreground/background messaging for browser
Defers work for browser to the background
* Test foreground port behaviors
* Inject foreground and background derived state services
* remove unnecessary class field
* Adhere to required options
* Add dderived state to CLI
* Prefer type definition in type parameters to options
* Prefer instance method
* Implements factory methods for common uses
* Remove nothing test
* Remove share subject reference
Share manages connector subjects internally and will reuse them until
refcount is 0 and the cleanup time has passed. Saving our own reference
just risks memory leaks without real testability benefits.
* Fix interaction state
* Add a small default time to limit timing failures
* Handle subscription race conditions
* Add Symbols to tracked emission types
This is a bit of a cheat, but Symbols can't be cloned, so
we need to nudge them to something we can handle.
They are rare enough that anyone hitting this is likely to
expect some special handling.
* Ref count state listeners to minimize storage activity
* Ensure statuses are updated
* Remove notes
* Use `test` when gramatically more proper
* Copy race and subscription improvements to single user
* Simplify observer initialization
* Correct parameter names
* Simplify update promises
test we don't accidentally deadlock along the `getFromState` path
* Fix save mock
* WIP: most tests working
* Avoid infinite update loop
* Avoid potential deadlocks with awaiting assigned promises
We were awaiting a promise assigned in a thenable. It turns out that
assignment occurs before all thenables are concatenated, which can cause
deadlocks. Likely, these were not showing up in tests because we're
using very quick memory storage.
* Fix update deadlock test
* Add user update tests
* Assert no double emit for multiple observers
* Add use intent to method name
* Ensure new subscriptions receive only newest data
TODO: is this worth doing for active user state?
* Remove unnecessary design requirement
We don't need to await an executing update promise, we
can support two emissions as long as the observable is
guaranteed to get the new data.
* Cleanup await spam
* test cleanup option behavior
* Remove unnecessary typecast
* Throw over coerce for definition options
* Fix state$ observable durability on cleanup
* Add a small default time to limit timing failures
* Handle subscription race conditions
* Add Symbols to tracked emission types
This is a bit of a cheat, but Symbols can't be cloned, so
we need to nudge them to something we can handle.
They are rare enough that anyone hitting this is likely to
expect some special handling.
* Ref count state listeners to minimize storage activity
* Ensure statuses are updated
* Remove notes
* Use `test` when gramatically more proper
* Copy race and subscription improvements to single user
* Simplify observer initialization
* Correct parameter names
* Simplify update promises
test we don't accidentally deadlock along the `getFromState` path
* Fix save mock
* WIP: most tests working
* Avoid infinite update loop
* Avoid potential deadlocks with awaiting assigned promises
We were awaiting a promise assigned in a thenable. It turns out that
assignment occurs before all thenables are concatenated, which can cause
deadlocks. Likely, these were not showing up in tests because we're
using very quick memory storage.
* Fix update deadlock test
* Add user update tests
* Assert no double emit for multiple observers
* Add use intent to method name
* Ensure new subscriptions receive only newest data
TODO: is this worth doing for active user state?
* Remove unnecessary design requirement
We don't need to await an executing update promise, we
can support two emissions as long as the observable is
guaranteed to get the new data.
* Cleanup await spam
* test cleanup option behavior
* Remove unnecessary typecast
* Throw over coerce for definition options
* Specify state provider for currently active user
* Split active and single user States
UserStateProvider is still the mechanism to build each State object.
The SingleUserState is basically a repeat of GlobalState, but with
additional scoping.
* Fixup global state cache
* fix fakers to new interface
* Make userId available in single user state
* Split providers by dependency requirements
This allows usage of the single state provider in contexts that would
otherwise form circular dependencies.
* Offer convenience wrapper classes for common use
* Import for docs
* Bind wrapped methods
* Handle switch messaging
TODO: handle loading state for account switcher
* Async updates required for state
* Fallback to email for current account avatar
* Await un-awaited promises
* Remove unnecessary Prune
Prune was getting confused in browser and deleting memory in browser on
account switch. This method isn't needed since logout already removes
memory data, which is the condition for pruning
* Fix temp password in browser
* Use direct memory access until data is serializable
Safari uses a different message object extraction than firefox/chrome
and is removing `UInt8Array`s. Until all data passed into StorageService
is guaranteed serializable, we need to use direct access in state
service
* Reload badge and context menu on switch
* Gracefully switch account as they log out.
* Maintain location on account switch
* Remove unused state definitions
* Prefer null for state
undefined can be misinterpreted to indicate a value has not been set.
* Hack: structured clone in memory storage
We are currently getting dead objects on account switch due to updating
the object in the foreground state service. However, the storage service
is owned by the background. This structured clone hack ensures that all
objects stored in memory are owned by the appropriate context
* Null check nullable values
active account can be null, so we should include null safety in the
equality
* Correct background->foreground switch command
* Already providing background memory storage
* Handle connection and clipboard on switch account
* Prefer strict equal
* Ensure structuredClone is available to jsdom
This is a deficiency in jsdom --
https://github.com/jsdom/jsdom/issues/3363 -- structured clone is well
supported.
* Fixup types in faker class
* Move Account Settings
* Add Another Test
* Refactor Tests
* Update Notification Bar to Get Value From Global
* Also Migrate Disable Context Menu
* Add Explanation
