4 ms·
RIPE's DNSSEC signer migration
- CaliforniaKarl 8y ago> When we introduced DNSSEC to all our zones back in 2005, we did this using some home-grown perl scripts. I am simultaneously cringing while nodding my head in approval. > In 2009, we switched from these perl scripts to a signer product developed by a company called Secure64. This is a dedicated signer solution that runs on Secure64's SourceT operating system, on HP servers with Itanium processors. I'm not going to malign them for choosing a proprietary product. Back when I did DNS admin for a company, I didn't trust myself to handle all the vagaries of DNSSEC, and so I relied on our Infoblox appliances to handle it, which they did seamlessly. Itanium, though… > We settled on standard Knot DNS for our signing solution… Interesting! I haven't heard of this before, and it looks interesting. Will keep in in mind if I ever end up needing to choose an authoritative DNS. > …and we have found the support from CZ.NIC to be really good. Very important. Helps keep the product alive. > …use of an HSM might actually be overkill. We have therefore decided to keep our setup simple, by adopting a design that uses a regular Linux server running Knot DNS, with keys on an encrypted disk partition. Hmmmmm, I wonder how they set that up! Just as a fun thing, I experimented with using a chain of systemd units to handle the unlocking and mounting of an encrypted EBS volume. It went like this: 1. A oneshot service which attached an EBS volume on start, and detached it on stop. 2. A mount unit that mounted the filesystem that contained the encrypted (wrapped) key. 3. A oneshot service which called KMS to "unwrap" the key into a RAM-based filesystem, and then call LUKS to unlock the encrypted filesystem. 4. A mount unit that mounted the unlocked filesystem. (Note that AWS Secrets Manager didn't exist at the time, so I couldn't use it. Also, I didn't want to rely on external infrastructure like a Vault.) It all worked because, instead of putting entries into `/etc/fstab`, systemd lets you create `mount` units with that can have the same dependency relationship as services. > Our old Secure64 signers do not export their private keys. This means that in order to migrate to the new signers, we will have to perform a KSK roll-over. Therefore, we will pre-publish DS records of KSKs from our new signers. A KSK rollover? That's always 'fun', and I wish them luck! Although, it'll be alot easier for them, compared to (for example) rolling over the root's KSK: See https://www.apnic.net/manage-ip/apnic-services/dnssec/keyroll https://www.apnic.net/manage-ip/apnic-services/dnssec/keyrol... and https://www.icann.org/news/announcement-2017-09-27-en https://www.icann.org/news/announcement-2017-09-27-en and https://www.icann.org/resources/press-material/release-2018-09-18-en https://www.icann.org/resources/press-material/release-2018-... which OMG just came out today (/me goes to post it).
- tptacek 8y agoThe private keys for DNSSEC signing must be stored safely. It is common to use a hardware security module (HSM) for this. However, this is not the only way to store keys safely. Keys can be generated and stored on encrypted disks, and if access to a server is restricted and limited to only certain functions, then it is acceptable to use such a setup. What? No, those are not comparable solutions at all. Rather, the latter solution, of relying on Linux FDE, defines the problem that HSMs solve away.