ripe-atlas-shrugd

Fake DNS resolver check using RIPE Atlas

Quickstart

You'll need the RIPE Atlas library:

$ pip3 install --user ripe.atlas.cousteau      # for creating measurements

You'll also probably need either the IANA root or the Yeti root:

$ wget -O iana-root.txt ftp://rs.internic.net/domain/named.root 
$ wget -O yeti-root.txt https://raw.githubusercontent.com/BII-Lab/Yeti-Project/master/domain/named.cache

Finally, you will need to set your RIPE Atlas key in the atlaskeys.py file.

You can then start a measurement using one of your root hints:

$ python3 shrugd-create.py iana-root.txt

Background & Overview

Currently (2015-Q4) the RIPE Atlas probes have two ways to perform DNS lookups:

1. They can do a lookup using the local resolver of the probe's network, or
2. The RIPE Atlas central server can perform DNS lookups before sending a measurement to the probe.

While this is satisfactory for most experiments, in some cases more control of the resolution process would be desirable. For example, the Yeti Project (https://yeti-dns.org/) uses an different set of root hints and needs to kick off resolution differently.

Fortunately, the RIPE Atlas probes are quite flexible and can send DNS packets with almost arbitrary contents. The approach in shrugd is to emulate the DNS resolution process by sending packets similar to the ones that a real DNS resolver would use.

Details

The software uses the RIPE Atlas streaming API. This returns the results from each measurement as they complete. In this way it performs somewhat like a real resolver in that packets are sent as needed and responses received asynchronously and processed in real time.

The basic approach is to start at the top of the DNS hierarchy (the root) and then progress down label by label until an answer for the question asked is received. Basically this works like this:

1. Send a query for the desired name to each of the IP addresses in the hints.
2. When the first reply for the current level completes, see if an answer is in the reply. If so, then we are done!
3. If the first reply contains no answer, but does contain an authority section with NS records, then get the addresses for those NS records and send a query for the desired name to each of those IP addresses. Goto step #2.

In step #3, we use the IP addresses listed in the additional section as the IP addresses for our NS servers, if they are present (this is glue). If there is no glue (for example if the NS servers are out-of-bailiwick), then shrugd looks up the IP addresses locally using the local DNS resolver. It is possible to perform full recursion to get this information, but this is a simplifying step and makes the resolution much more straightforward.

This process performs a superset of what a normal resolver will do. Firstly, a resolver will typically not send the query to all authority servers at a given level of the hierarchy. Secondly, a real resolver will typically have most of the information cached very quickly - certainly the root and TLD name servers. Thirdly, a real resolver will use RTT information about IP addresses to prefer specific servers.

We use the superset since we can't know the cache behavior. What we can do is look at the full set of results and make some analysis about possible resolver behavior, setting lower-bounds and upper-bounds on performance, and making guesses about expected results.

Right now probe selection is out of scope, but the measurement can be conducted separately on different probes as desired.