5984,6984 - Pentesting CouchDB
Basic Information
CouchDB is a document-oriented database and within each document fields are stored as key-value maps. Fields can be either a simple key/value pair, list, or map.
Each document that is stored in the database is given a document-level unique identifier (_id
) as well as a revision (_rev
) number for each change that is made and saved to the database.
Default port: 5984(http), 6984(https)
Automatic Enumeration
Manual Enumeration
Banner
This issues a GET request to installed CouchDB instance. The reply should look something like on of the following:
Note that if accessing the root of couchdb you receive a 401 Unauthorized
with something like this: {"error":"unauthorized","reason":"Authentication required."}
you won't be able to access the banner or any other endpoint.
Info Enumeration
These are the endpoints where you can access with a GET request and extract some interesting info. You can find more endpoints and more detailed descriptions in the couchdb documentation.
/_active_tasks
List of running tasks, including the task type, name, status and process ID./_all_dbs
Returns a list of all the databases in the CouchDB instance./_cluster_setup
Returns the status of the node or cluster, per the cluster setup wizard./_db_updates
Returns a list of all database events in the CouchDB instance. The existence of the_global_changes
database is required to use this endpoint./_membership
Displays the nodes that are part of the cluster ascluster_nodes
. The fieldall_nodes
displays all nodes this node knows about, including the ones that are part of the cluster./_scheduler/jobs
List of replication jobs. Each job description will include source and target information, replication id, a history of recent event, and a few other things./_scheduler/docs
List of replication document states. Includes information about all the documents, even incompleted
andfailed
states. For each document it returns the document ID, the database, the replication ID, source and target, and other information./_scheduler/docs/{replicator_db}
/_scheduler/docs/{replicator_db}/{docid}
/_node/{node-name}
The/_node/{node-name}
endpoint can be used to confirm the Erlang node name of the server that processes the request. This is most useful when accessing/_node/_local
to retrieve this information./_node/{node-name}/_stats
The_stats
resource returns a JSON object containing the statistics for the running server. The literal string_local
serves as an alias for the local node name, so for all stats URLs,{node-name}
may be replaced with_local
, to interact with the local node’s statistics./_node/{node-name}/_system
The _systemresource returns a JSON object containing various system-level statistics for the running server. You can use ___local
as {node-name} to get current node info./_node/{node-name}/_restart
/_up
Confirms that the server is up, running, and ready to respond to requests. Ifmaintenance_mode
istrue
ornolb
, the endpoint will return a 404 response./_uuids
Requests one or more Universally Unique Identifiers (UUIDs) from the CouchDB instance./_reshard
Returns a count of completed, failed, running, stopped, and total jobs along with the state of resharding on the cluster.
More interesting information can be extracted as explained here: https://lzone.de/cheat-sheet/CouchDB
Database List
If that request responds with a 401 unauthorised, then you need some valid credentials to access the database:
In order to find valid Credentials you could try to bruteforce the service.
This is an example of a couchdb response when you have enough privileges to list databases (It's just a list of dbs):
Database Info
You can obtain some database info (like number of files and sizes) accessing the database name:
Document List
List each entry inside a database
Read Document
Read the content of a document inside a database:
CouchDB Privilege Escalation CVE-2017-12635
Thanks to the differences between Erlang and JavaScript JSON parsers you could create an admin user with credentials hacktricks:hacktricks
with the following request:
****More information about this vuln here.
CouchDB RCE
Erlang Cookie
In the CouchDB docs, in the cluster set-up section, it talks about the different ports used by CouchDB:
CouchDB in cluster mode uses the port
5984
just as standalone, but it also uses5986
for node-local APIs.Erlang uses TCP port
4369
(EPMD) to find other nodes, so all servers must be able to speak to each other on this port. In an Erlang Cluster, all nodes are connected to all other nodes. A mesh.
And then there’s an interesting warning:
If we look in the process list, we can see that cookie, “monster”:
You can read this section to learn how to abuse Erlangs cookies to obtain RCE. Also, you can read some Canape HTB machine writeup like this one to see and practice how to exploit this vuln.
Successful CVE-2018-8007 with local.ini write permissions
In writing this post, I found a new CVE had been released for CouchDB from mdsec, CVE-2018-8007. It also requires writes to the local.ini
file, so it isn’t a useful option for Canape. But since I’ve already made it writable as root, let’s see if we can get it to work.
Start with a clean and now writable local.ini
(and a backup):
We can use curl to modify the origins in the local.ini
file. The vulnerability here is that if we use curl to put a new origin and then newlines, we can write additional stuff, including a new header and details. So we’ll take advantage of the [os_daemons]
field, and add a process for CouchDB to try to keep running:
In the root shell, we can see what changes:
And yet, the file isn’t there:
If we look at the processes running with “couchdb” in the cmdline, we see not only the line command line that gives us the cookie value we used earlier, but also runsrv couchdb
:
If we kill that process, it comes right back (notice the new pid):
And, on restart, runs the OS_Daemons:
Successful Attempt Via CVE-2017-12636 with local.ini write permissions
CVE-2017-12636 allows for code execution through the couchdb process. However, it won’t work in this configuration.
There are a few POCs out there as reference:
We’d need to write a new query_server, and then invoke that. When Canape was released, most of the POCs were for couchdb 1.x, but this box is running 2, so the query_servers path from most of the POCs doesn’t exist. That’s changed now, but we’ll walk the same steps. First, get the version, and show that the 1.X path doesn’t exist:
Update with the new path for 2.0:
From there, we should add a query_server and then invoke it, but we aren’t able to.
Some Googling shows that this is an issue with permissions. In fact, if we check with out root shell, we can see that the local.ini
file is not writable by anyone, let alone www-data:
So that’s a dead end for Canape. But if we want to try to get it working, we can make it readable with our root or homer access, and continue down this path. We’ll make a backup of the original so we can see what changes:
Now, back to our www-data shell:
We get back the previous value for the cmd query server, which means success. And in the root shell, we can see it worked:
Now, we should be able to create a db, and then a document in that db, and the request it with a view that maps our query_server to get execution.
Create db and document:
Request it in a view:
Summary with a different payload
Shodan
port:5984 couchdb
References
Last updated