[
https://issues.apache.org/jira/browse/SOLR-12514?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16523546#comment-16523546
]
Jan Høydahl commented on SOLR-12514:
------------------------------------
There are a bunch of complexities arising here, so it's tempting to throw a 302
as a quick-fix. We can then advertise that clients using this type of auth can
either comply with 302 responses or preemptively read the cluster state and
route requests intelligently to nodes hosting a collection - the same way as
CloudSolrClient already does.
If we go down the proxy style forwarding route, I imagine that we get other
problems down the road. Say we add a IP-based auth plugin and tjen the request
gets forwarded to another node, and IP auth fails etc.
On the other hand - if we want Admin UI to play nicely with an authenticated
user (given that the UI gets auth support in the future), we end up with CORS
issues if the UI has to send query requests to different nodes...
> Rule-base Authorization plugin skips authorization if querying node does not
> have collection replica
> ----------------------------------------------------------------------------------------------------
>
> Key: SOLR-12514
> URL: https://issues.apache.org/jira/browse/SOLR-12514
> Project: Solr
> Issue Type: Bug
> Security Level: Public(Default Security Level. Issues are Public)
> Components: security
> Affects Versions: 7.3.1
> Reporter: Mahesh Kumar Vasanthu Somashekar
> Priority: Major
> Attachments: SOLR-12514.patch, Screen Shot 2018-06-24 at 9.36.45
> PM.png, security.json
>
>
> Solr serves client requests going throught 3 steps - init(), authorize() and
> handle-request ([link
> git-link|https://github.com/apache/lucene-solr/blob/releases/lucene-solr/7.3.1/solr/core/src/java/org/apache/solr/servlet/HttpSolrCall.java#L471]).
> init() initializes all required information to be used by authorize().
> init() skips initializing if request is to be served remotely, which leads to
> skipping authorization step ([link
> git-link|https://github.com/apache/lucene-solr/blob/releases/lucene-solr/7.3.1/solr/core/src/java/org/apache/solr/servlet/HttpSolrCall.java#L291]).
> init() relies on 'cores' object which only has information of local node
> (which is perfect as per design). It should actually be getting security
> information (security.json) from zookeeper, which has global view of the
> cluster.
>
> Example:
> SolrCloud setup consists of 2 nodes (solr-7.3.1):
> live_nodes: [
> "localhost:8983_solr",
> "localhost:8984_solr",
> ]
> Two collections are created - 'collection-rf-1' with RF=1 and
> 'collection-rf-2' with RF=2.
> Two users are created - 'collection-rf-1-user' and 'collection-rf-2-user'.
> Security configuration is as below (security.json attached):
> "authorization":{
> "class":"solr.RuleBasedAuthorizationPlugin",
> "permissions":[
> { "name":"read", "collection":"collection-rf-2", "role":"collection-rf-2",
> "index":1}
> ,
> { "name":"read", "collection":"collection-rf-1", "role":"collection-rf-1",
> "index":2}
> ,
> { "name":"read", "role":"*", "index":3}
> ,
> ...
> "user-role":
> { "collection-rf-1-user":[ "collection-rf-1"], "collection-rf-2-user":[
> "collection-rf-2"]}
> ,
> ...
>
> Basically, its setup to that 'collection-rf-1-user' user can only access
> 'collection-rf-1' collection and 'collection-rf-2-user' user can only access
> 'collection-rf-2' collection.
> Also note that 'collection-rf-1' collection replica is only on
> 'localhost:8983_solr' node, whereas ''collection-rf-2' collection replica is
> on both live nodes.
>
> Authorization does not work as expected for 'collection-rf-1' collection:
> $ curl -u collection-rf-2-user:password
> 'http://*localhost:8983*/solr/collection-rf-1/select?q=*:*'
> <html>
> <head>
> <meta http-equiv="Content-Type" content="text/html;charset=utf-8"/>
> <title>Error 403 Unauthorized request, Response code: 403</title>
> </head>
> <body><h2>HTTP ERROR 403</h2>
> <p>Problem accessing /solr/collection-rf-1/select. Reason:
> <pre> Unauthorized request, Response code: 403</pre></p>
> </body>
> </html>
> $ curl -u collection-rf-2-user:password
> 'http://*localhost:8984*/solr/collection-rf-1/select?q=*:*'
> {
> "responseHeader":{
> "zkConnected":true,
> "status":0,
> "QTime":0,
> "params":{
> "q":"*:*"}},
> "response":{"numFound":0,"start":0,"docs":[]
> }}
>
> Whereas authorization works perfectly for 'collection-rf-2' collection (as
> both nodes have replica):
> $ curl -u collection-rf-1-user:password
> 'http://*localhost:8984*/solr/collection-rf-2/select?q=*:*'
> <html>
> <head>
> <meta http-equiv="Content-Type" content="text/html;charset=utf-8"/>
> <title>Error 403 Unauthorized request, Response code: 403</title>
> </head>
> <body><h2>HTTP ERROR 403</h2>
> <p>Problem accessing /solr/collection-rf-2/select. Reason:
> <pre> Unauthorized request, Response code: 403</pre></p>
> </body>
> </html>
> $ curl -u collection-rf-1-user:password
> 'http://*localhost:8983*/solr/collection-rf-2/select?q=*:*'
> <html>
> <head>
> <meta http-equiv="Content-Type" content="text/html;charset=utf-8"/>
> <title>Error 403 Unauthorized request, Response code: 403</title>
> </head>
> <body><h2>HTTP ERROR 403</h2>
> <p>Problem accessing /solr/collection-rf-2/select. Reason:
> <pre> Unauthorized request, Response code: 403</pre></p>
> </body>
> </html>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
