Mailing-List: contact dev-help@couchdb.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@couchdb.apache.org
Date: Sun, 6 Apr 2014 18:20:15 +0000 (UTC)
From: "Robert Newson (JIRA)" <jira@apache.org>
To: dev@couchdb.apache.org
Message-ID: <JIRA.12706810.1396738933146.71342.1396808415455@arcas>
In-Reply-To: <JIRA.12706810.1396738933146@arcas>
References: <JIRA.12706810.1396738933146@arcas>
Subject: [jira] [Commented] (COUCHDB-2221) malformed iterations field in
 _users doc causes authentication hang
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/COUCHDB-2221?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13961479#comment-13961479 ] 

Robert Newson commented on COUCHDB-2221:
----------------------------------------

I pondered this some more and it bothers me that users have free reign over how many iterations they use. The server administrator should be in charge here. Thus I propose https://gist.github.com/rnewson/6f6cc38e9f5e9ac22e65 which will allow couchdb to reject (with a 403) any user doc using either a too weak or too strong iteration count, defaulting to generous bounds. The guard clause on verify_iterations also serves to prevent the infinite loop this ticket was raised about.

Let's regroup Monday to decide what combination of this we want to go with and which release besides 1.6.x it might go in.

> malformed iterations field in _users doc causes authentication hang
> -------------------------------------------------------------------
>
>                 Key: COUCHDB-2221
>                 URL: https://issues.apache.org/jira/browse/COUCHDB-2221
>             Project: CouchDB
>          Issue Type: Bug
>      Security Level: public(Regular issues) 
>          Components: Database Core
>            Reporter: Isaac Z. Schlueter
>            Assignee: Joan Touzet
>
> Create a user account with the following details:
> {
>   "_id":"org.couchdb.user:test-user",
>   "name":"test-user",
>   "password":"this is a test"
>   "roles":[],
>   "type":"user"
> }
> CouchDB will PBKDF2-ify the password in the _users doc.  So far so good.
> Then, try this:
> ubuntu@ip-172-31-35-228:~$ curl http://localhost:5984/_users/org.couchdb.user:test-user -u "test-user:this is not the correct password" -vvv
> * About to connect() to localhost port 5984 (#0)
> *   Trying 127.0.0.1... connected
> * Server auth using Basic with user 'test-user'
> > GET /_users/org.couchdb.user:test-user HTTP/1.1
> > Authorization: Basic dGVzdHVzZXI6dGhpcyBpcyBub3QgdGhlIGNvcnJlY3QgcGFzc3dvcmQ=
> > User-Agent: curl/7.22.0 (x86_64-pc-linux-gnu) libcurl/7.22.0 OpenSSL/1.0.1 zlib/1.2.3.4 libidn/1.23 librtmp/2.3
> > Host: localhost:15984
> > Accept: */*
> >
> And then it hangs indefinitely.
> This does not happen when the user account uses password_sha.  For example:
> ubuntu@ip-172-31-35-228:~$ curl http://localhost:15984/_users/org.couchdb.user:testuserasdf -u "testuserasdf:this is not the correct password" -vvv
> * About to connect() to localhost port 15984 (#0)
> *   Trying 127.0.0.1... connected
> * Server auth using Basic with user 'testuserasdf'
> > GET /_users/org.couchdb.user:testuserasdf HTTP/1.1
> > Authorization: Basic dGVzdHVzZXJhc2RmOnRoaXMgaXMgbm90IHRoZSBjb3JyZWN0IHBhc3N3b3Jk
> > User-Agent: curl/7.22.0 (x86_64-pc-linux-gnu) libcurl/7.22.0 OpenSSL/1.0.1 zlib/1.2.3.4 libidn/1.23 librtmp/2.3
> > Host: localhost:15984
> > Accept: */*
> >
> < HTTP/1.1 401 Unauthorized
> < Server: CouchDB/1.5.0 (Erlang OTP/R14B04)
> < Date: Sat, 05 Apr 2014 22:58:54 GMT
> < Content-Type: text/plain; charset=utf-8
> < Content-Length: 67
> < Cache-Control: must-revalidate
> <
> {"error":"unauthorized","reason":"Name or password is incorrect."}
> * Connection #0 to host localhost left intact
> * Closing connection #0
> This is a serious and urgent problem for npm.
> At the urging of many people in the CouchDB and Node.js community, we've been migrating users to pbkdf2 accounts.  However, rather than quickly report authorization failures, it hangs indefinitely, and eventually our TLS terminator returns a 500 or our CDN returns a 503.
> Because the appropriate HTTP response code is not being returned, we cannot hope to properly handle the situation.  It looks like the server has just fallen over.  Already the user experience has started to get pretty awful.
> What's worse, I fear that this is a DOS exploit, because it ties up a connection for a very long time.  The npm registry is somewhat insulated by our CDN, but any CouchDB using pbkdf2 password storage is vulnerable.


--
This message was sent by Atlassian JIRA
(v6.2#6252)