This commit adds a test to ensure that all request handlers are wrapped
by ProfileSecuredRequestHandler.
Original commit: elastic/x-pack-elasticsearch@26473d0ddc
Currently we copy the authorization header from every rest request to the action request. This is not
necessary because the user associated with each request is copied into the context and then if the
request leaves the node, the user will be serialized into a string and attached as a header.
This commit removes the copying of the authorization header as it is not necessary and by not copying
it, we limit the amount of copies we make of this sensitive information.
Original commit: elastic/x-pack-elasticsearch@4e5ba4b4aa
This commit adds document and field level security to Shield.
Field level security can be enabled by adding the `fields` option to a role in the `role.yml` file.
For example:
```yaml
customer_care:
indices:
'*':
privileges: read
fields:
- issue_id
- description
- customer_handle
- customer_email
- customer_address
- customer_phone
```
The `fields` list is an inclusive list of fields that controls what fields should be accessible for that role. By default all meta fields (_uid, _type, _source, _ttl etc) are also included, otherwise ES or specific features stop working. The `_all` field if configured, isn't included by default, since that actually contains data from all the other fields. If the `_all` field is required then this needs to be added to the `fields` list in a role. In the case of the content of the `_source` field and `_field_names` there is special filtering in place so that only the content relevant for the role are being returned.
If no `fields` is specified then field level security is disabled for that role and all fields in an index are accessible.
Field level security can be setup per index group.
Field level security is implemented at the Lucene level by wrapping a directory index reader and hides fields away that aren't in the `field` list defined with the role of the current user. It as if the other fields never existed.
* Any `realtime` read operation from the translog is disabled. Instead this operations fall back to the Lucene index, which makes these operations compatible with field level security, but there aren't realtime.
* If user with role A executes first and the result gets cached and then a user with role B executes the same query results from the query executed with role A would be returned. This is bad and therefore the query cache is disabled.
* For the same reason the request cache is also disabled.
* The update API is blocked. An update request needs to be executed via a role that doesn't have field level security enabled.
Document level security can be enabled by adding the `query` option to a role in the `role.yml` file:
```yaml
customer_care:
indices:
'*':
privileges: read
query:
term:
department_id: 12
```
Document level security is implemented as a filter that filters out documents there don't match with the query. This is like index aliases, but better, because the role query is embedded on the lowest level possible in ES (Engine level) and on all places the acquire an IndexSearcher the role query will always be included. While alias filters are applied at a higher level (after the searcher has been acquired)
Document level security can be setup per index group.
Right now like alias filters the document level security isn't applied on all APIs. Like for example the get api, term vector api, which ignore the alias filter. These apis do acquire an IndexSearcher, but don't use the IndexSearcher itself and directly use the index reader to access the inverted index and there for bypassing the role query. If it is required to these apis need document level security too the the implementation for document level security needs to change.
Closeselastic/elasticsearch#341
Original commit: elastic/x-pack-elasticsearch@fac085dca6
PreProcessModule was an alternate way to customize another module's
behavior inside plugins. The preferred (and only in the future) way to
do this is with onModule in the plugin itself. This change moves the
only two remaining users of PreProcessModule to do so in their
respective plugins. The use case was adding roles for shield
authorization, but these roles were really static, so there was no
reason they could not be configured up front.
Original commit: elastic/x-pack-elasticsearch@e67ac2dcb6
This adds the extension points necessary to enable a user to write a elasticsearch plugin
that can integrate with Shield and add a custom authentication realm. For the most part,
the work here just exposes the existing interfaces we have been using for Realms and
factories to create realms. An additional interface was added to allow for a custom
authentication failure handler to be used. This was needed to support use cases like SSO
and Kerberos where additional headers may need to be sent to the user or a different
HTTP response code would need to be sent.
Relates to elastic/elasticsearch#24
Original commit: elastic/x-pack-elasticsearch@13442e5919
SpawnModules will be going away very soon as part of
elastic/elasticsearchelastic/elasticsearch#12783. This change removes its use from all
x-plugins.
Most spawnmodules uses here were to either collect a number of modules
into one (so the modules were just moved up into the plugin itself), or
to spawn a module which interacted with an extension point from ES. This
change moves those, as well as most uses of PreProcessModule, to use
onModule.
Original commit: elastic/x-pack-elasticsearch@6430e35379
Today the XContent building of the response for the ClearRealmsCacheResponse is broken and causes
an exception to be thrown. This fixes the building of the response and adds tests that call the HTTP
endpoint and do a basic check on the response.
Closeselastic/elasticsearch#390
Original commit: elastic/x-pack-elasticsearch@8ad9dae4ea
This commit changes the groupId to the above mentioned one
so that S3 uploads will end up in the right bucket. This will
allow the Elasticsearch plugin manager to install the commercial
plugins like
```
bin/plugin install {watcher,shield,license,marvel}
```
like the official ones.
Original commit: elastic/x-pack-elasticsearch@642f1f006a
We can just run these during the integration test phase: there is
no benefit in running them during `mvn test` too.
Original commit: elastic/x-pack-elasticsearch@4b275920e2
With elastic/elasticsearchelastic/elasticsearch#12623 base test classes were renamed
to use "TestCase" suffix. This updates x-plugins to reflect those
name changes. It also renames some tests that were marked
with @Slow (which was forbidden with elastic/elasticsearchelastic/elasticsearch#12617 and
elastic/elasticsearch elastic/elasticsearch#12618) to use the IT suffix to run
under `mvn verify`.
Original commit: elastic/x-pack-elasticsearch@05ffe2f202
homeFile() is removed and should not be used, we need to cleanup,
but this is just a rote change to get builds green.
Original commit: elastic/x-pack-elasticsearch@05d0fb4a7c
Prior to this commit, we were InetAddress.getLocalHost() to get the hostname and host
address when auditing. This is different than how we report the node's hostname and host
address in other places where we use NetworkUtils. This caused false failures to be seen
with the IndexAuditTrail tests. This commit switches the audit trails to use the NetworkUtils
methods.
Closeselastic/elasticsearch#285
Original commit: elastic/x-pack-elasticsearch@c0bd7e94f6
Today, if a LDAP server is down and the LDAP realm uses the user search mechanism this will prevent the
node from starting up. This is not ideal because users can still authenticate with another realm if it is
configured. This change tries to create the connection pool on initialization but if it fails, creation will retried
on each attempted authentication until the server is available again.
Closeselastic/elasticsearch#107
Original commit: elastic/x-pack-elasticsearch@f2ccf858ff
Currently, we use the local cluster state when determining if the index audit trail can be
started. This is correct when we are logging to the same cluster but is incorrect when we
log to a remote cluster. Instead we should try to initialize the client and get the remote
cluster's state.
This also changes the enqueue method to stop throwing an exception on failing to add a
message to the queue. The exception was unnecessary and causing hard to read logs.
It is now replaced with a simple warn log message.
Closeselastic/elasticsearch#317
Original commit: elastic/x-pack-elasticsearch@238e9159b3
Currently, we attach the internal audit user to all requests. This is incorrect for requests that
need to be sent to a remote cluster. For these cases, we should require a user to be defined
to access the remote cluster if it is protected by Shield.
Additionally, the origin_address field for rest request fields is formatted differently than other
address fields. This changes the field to only be the remote address.
Closeselastic/elasticsearch#278Closeselastic/elasticsearch#279
Original commit: elastic/x-pack-elasticsearch@a5f86b1974
Martijn van Groningen
Areek Zillur
Britta Weber
jaymode
Jason Tedor
David Pilato
Boaz Leskes
Simon Willnauer
Ryan Ernst
Adrien Grand
uboness
Alexander Reelsen
Robert Muir
Tanguy Leroux