* Allow more complex errors from plugins
This enables more complex types to be registered and returned from plugins.
* Register common error types
This is a slightly less drastic change, which keeps the HTTPCodedError
as an interface.
* Remove replication error from list
* porting identity to OSS
* changes that glue things together
* add testing bits
* wrapped entity id
* fix mount error
* some more changes to core
* fix storagepacker tests
* fix some more tests
* fix mount tests
* fix http mount tests
* audit changes for identity
* remove upgrade structs on the oss side
* added go-memdb to vendor
* disable raw endpoint by default
* adding docs
* config option raw -> raw_storage_endpoint
* docs updates
* adding listing on raw endpoint
* reworking tests for enabled raw endpoints
* root protecting base raw endpoint
* Lazy load plugins to avoid setup-unwrap cycle
* Remove commented blocks
* Refactor NewTestCluster, use single core cluster on basic plugin tests
* Set c.pluginDirectory in TestAddTestPlugin for setupPluginCatalog to work properly
* Add special path to mock plugin
* Move ensureCoresSealed to vault/testing.go
* Use same method for EnsureCoresSealed and Cleanup
* Bump ensureCoresSealed timeout to 60s
* Correctly handle nil opts on NewTestCluster
* Add metadata flag to APIClientMeta, use meta-enabled plugin when mounting to bootstrap
* Check metadata flag directly on the plugin process
* Plumb isMetadataMode down to PluginRunner
* Add NOOP shims when running in metadata mode
* Remove unused flag from the APIMetadata object
* Remove setupSecretPlugins and setupCredentialPlugins functions
* Move when we setup rollback manager to after the plugins are initialized
* Fix tests
* Fix merge issue
* start rollback manager after the credential setup
* Add guards against running certain client and server functions while in metadata mode
* Call initialize once a plugin is loaded on the fly
* Add more tests, update basic secret/auth plugin tests to trigger lazy loading
* Skip mount if plugin removed from catalog
* Fixup
* Remove commented line on LookupPlugin
* Fail on mount operation if plugin is re-added to catalog and mount is on existing path
* Check type and special paths on startBackend
* Fix merge conflicts
* Refactor PluginRunner run methods to use runCommon, fix TestSystemBackend_Plugin_auth
Previously we lowercased names on ingress but not on lookup or delete
which could cause unexpected results. Now, just unilaterally lowercase
policy names on write and delete. On get, to avoid the performance hit
of always lowercasing when not necessary since it's in the critical
path, we have a minor optimization -- we check the LRU first before
normalizing. For tokens, because they're already normalized when adding
policies during creation, this should always work; it might just be
slower for API calls.
Fixes#3187
* Add automatic plugin reload
* Refactor builtin/backend
* Remove plugin reload at the core level
* Refactor plugin tests
* Add auto-reload test case
* Change backend to use sync.RWMutex, fix dangling test plugin processes
* Add a canary to plugin backends to avoid reloading many times (#3174)
* Call setupPluginCatalog before mount-related operations in postUnseal
* Don't create multiple system backends since core only holds a reference (#3176)
to one.
* Add plugin reload capability on all mounts for a specific plugin type
* Comments cleanup
* Add per-mount plugin backend reload, add tests
* Fix typos
* Remove old comment
* Reuse existing storage view in reloadPluginCommon
* Correctly handle reloading auth plugin backends
* Update path to plugin/backend/reload
* Use multierrors on reloadMatchingPluginMounts, attempt to reload all mounts provided
* Use internal value as check to ensure plugin backend reload
* Remove connection state from request for plugins at the moment
* Minor cleanup
* Refactor tests
* Store original request path in WrapInfo as CreationPath
* Add wrapping_token_creation_path to CLI output
* Add CreationPath to AuditResponseWrapInfo
* Fix tests
* Add and fix tests, update API docs with new sample responses
We still perform validation on the token, so if the call makes it
through to this endpoint it's got a valid token (either explicitly
specified in data or as the request token). But this allows
introspection for sanity/safety checking without revoking the token in
the process.
gRPC doesn't have a handler for recovering from a panic like a normal
HTTP request so a panic will actually kill Vault's listener. This
basically copies the net/http logic for managing this.
The SSH-specific logic is removed here as the underlying issue is caused
by the request forwarding mechanism.
