Implementing DNSSEC for a couple of years now while playing with many different DNS options such as TTL values, I came around an error message from DNSViz pointing to possible problems when the TTL of a signed resource record is longer than the lifetime of the DNSSEC signature itself. Since I was not fully aware of this (and because I did not run into a real error over the last years) I wanted to test it more precisely.
In my last blogpost I showed how to perform a DNSSEC KSK rollover. I did it quite slowly and carefully. This time I am looking into an emergency rollover of the KSK. That is: What to do if your KSK is compromised and you must replace it IMMEDIATELY.
I am listing the procedures and commands I used to replace the KSK of my delegated subdomain dyn.weberdns.de with BIND. And as you might already suggest it, I am showing DNSViz graphs after every step since it greatly reveals the current DNSKEYs etc.
Probably the most crucial part in a DNSSEC environment is the maintenance of the key-signing key, the KSK. You should rollover this key on a regular basis, though not that often as the zone signing keys, the ZSKs. I am doing a KSK rollover every 2 years.
In the following I will describe the two existing methods for a KSK rollover along with a step-by-step guide how I performed such a rollover for my zone “weberdns.de”. Of course again with many graphics from DNSViz (with “redundant edges”) that easily reveal the keys and signatures at a glance.
Note that this blogpost is NOT about the Root Zone KSK Rollover that appears in 2017/2018. It is merely about your OWN zone that is secured via DNSSEC.
If you are already familiar with DNSSEC this is quite easy: How to sign a delegated subdomain zone. For the sake of completeness I am showing how to generate and use the appropriate DS record in order to preserve the chain of trust for DNSSEC.
Until now I generated all SSHFP resource records on the SSH destination server itself via ssh-keygen -r <name>. This is quite easy when you already have an SSH connection to a standard Linux system. But when connecting to third party products such as routers, firewalls, whatever appliances, you don’t have this option. Hence I searched and found a way to generate SSHFP resource records remotely. Here we go:
This is actually a bad user experience problem: To generally omit the manual verification of SSH key fingerprints I am using SSHFP. With fully qualified domain names (FQDN) as the hostname for SSH connections such as ssh nb10.weberlab.de this works perfectly. However, admins are lazy and only use the hostname without the domain suffix to connect to their servers since the domain search does the rest: ssh nb10. Not so for SSHFP which fails since the default OpenSSH client does not use canonicalization for its DNS queries. Hence you must explicitly enable canonicalization for OpenSSH.
I am intensely using the SSH Public Key Fingerprint (SSHFP, RFC 4255) in all of my environments. Since my zones are secured via DNSSEC I got rid of any “authenticity of host ‘xyz’ can’t be established” problems. As long as I am using my central jump host with OpenSSH and the “VerifyHostKeyDNS yes” option I can securely login into any of my servers without any warnings. Great!
However, I encountered a couple of daily problems when using SSHFP. One of them was the question whether SSHFP works behind CNAMEs, that is, when connecting to an alias. Short answer: yes. Some more details here:
What is the biggest problem of PGP? The key distribution. This is well-known and not new at all. What is new is the OPENPGPKEY DNS resource record that delivers PGP public keys for mail addresses. If signed and verified with DNSSEC a mail sender can get the correct public key for his recipient. This solves both key distribution problems: 1) the delivery of the public key and 2) the authenticity of the key itself, i.e., that you’re using the correct key to encrypt a mail.
The “DNS-Based Authentication of Named Entities (DANE) Bindings for OpenPGP” is specified in the experimental RFC 7929. Let’s have a look on how you can add your public key into the zone file of your DNS server.
I really like the kind of security features that are easy to use. The CAA “DNS Certification Authority Authorization” is one of those. As a domain administrator you must only generate the appropriate CAA records and you’re done. (Unlike other security features such as HPKP that requires deep and careful planning or DANE which is not used widely.) Since the check of CAA records is mandatory for CAs since 8. September 2017, the usage of those records is quite useful because it adds another layer of security.
I know that BIND correctly changes the serial numbers of zones when it is enabled with inline signing and auto-dnssec. However, I got confused one more time as I looked on some of my SOA records. So, just for the record, here is an example how the serial numbers increase while the admin has not changed anything manually on the zone files.
The usage of the SSHFP resource record helps admins to authenticate the SSH server before they are exposing their credentials or before a man-in-the-middle attack occurs. This is only one great extension of DNSSEC (beside DANE whose TLSA records can be used to authenticate HTTPS/SMTPS servers).
While there are some great online tools for checking the mere DNS (1, 2), the correct DNSSEC signing (3, 4), or the placement of TLSA resource records for DANE (5, 6, 7), I have not found an online SSHFP validator. That’s the idea:
It is quite common that organizations use some kind of TLS decryption to have a look at the client traffic in order to protect against malware or evasion. (Some synonyms are SSL/TLS interception, decryption, visibility, man-in-the-middle, …) Next-generation firewalls as well as proxies implement such techniques, e.g., Palo Alto Networks or Blue Coat. To omit the certificate warnings by the clients, all spoofed certificates are signed by an internal root CA that is known to all internal clients. For example, the root CA is published via group policies to all end nodes.
But what happens if the DNS-based Authentication of Named Entities (DANE) is widely used within browsers? From the CA perspective, the spoofed certificates are valid, but not from the DANE perspective. To my mind we need something like an on-the-fly TLSA record spoofing technique that works in conjunction with TLS decryption.
After the implementation of DNS and DNSSEC (see the last posts) it is good to do some reconnaissance attacks against the own DNS servers. Especially to see the NSEC or NSEC3 differences, i.e., whether zone walking (enumeration) is feasible or not.
For many different kinds of DNS reconnaissance the tool dnsrecon can be used. In this post I will focus on the -z option which is used for DNSSEC zone walking, i.e., walk leaf by leaf of the whole DNS zone.
By default DNSSEC uses the next secure (NSEC) resource record “to provide authenticated denial of existence for DNS data”, RFC 4034. This feature creates a complete chain of all resource records of a complete zone. While it has its usage to prove that no entry exists between two other entries, it can be used to “walk” through a complete zone, known as zone enumeration. That is: an attacker can easily gather all information about a complete zone by just using the designed features of DNSSEC.
For this reason NSEC3 was introduced: It constructs a chain of hashed and not of plain text resource records (RFC 5155). With NSEC3 enabled it is not feasible anymore to enumerate the zone. The standard uses a hash function and adds the NSEC3PARAM resource record to the zone which provides some details such as the salt.
One important maintenance requirement for DNSSEC is the key rollover of the zone signing key (ZSK). With this procedure a new public/private key pair is used for signing the resource records, of course without any problems for the end user, i.e., no falsified signatures, etc.
In fact it is really simply to rollover the ZSK with BIND. It is almost one single CLI command to generate a new key with certain time ranges. BIND will use the correct keys at the appropriate time automatically. Here we go: