DNS name resolution queries can be secured by DNSSEC to avoid various spoofing attacks. When a local validating DNS resolver is in use, all software can potentially benefit from local DNSSEC validation if the system is configured properly. The root zone provides a global trust anchor that in turn allows for validation of DNS records in signed zones. Other trust anchors can be configured to explicitly protect known DNS subtrees.
Applications that don't understand DNSSEC are transparently protected by the local validating resolver which reports name resolution failure whenever validation of a DNS record fails. On the other hand, applications that know about DNSSEC can distinguish validated DNS records from DNS records in unsigned zones. Such applications can use DNSSEC validated data for example to initiate TLS sessions. A TLS library can do that for the application.
Client side DNSSEC deployment
DNSSEC brings data integrity and authenticity to plain DNS. This is very important, since one can publish different cryptographic (or any trusted) data in DNS database. This enables new types of applications and functionality like:
- Verify a remote server SSH fingerprint using SSHFP record in DNS (RFC 4255)
- Verify a TLS certificate offered by the remote server using DANE and TLSA record in DNS (RFC 6698)
- Get IPsec keys for a particular remote host automatically using IPSECKEY record in DNS (RFC 4025)
- Get X.509 or OpenPGP certificates using CERT record in DNS (RFC 4398)
- Verify that a specific Certification Authority is authorized to issue a certificate for a particular domain, using the CAA record in DNS (RFC 6844)
Validating resolver can validate the signed DNS response and set AD flag saying that it did the validation. One may not trust the network-provided DNS resolver and/or the channel (network) between the host and resolver.
There are currently two meaningful approaches for implementing DNSSEC aware applications:
- Use some intelligent name resolution library, which is able to do the validation (e.g. getdns library)
- Use locally running validating resolver (e.g. Unbound) and trust the AD flag in DNS response
This section explains the solution based on local validating resolver. This approach is beneficial also for applications that are not DNSSEC aware, since these will always get only responses that were validated. This improves the overall security of the system and of all applications that are doing domain name resolution for any purpose.
You can find more information in the Writing an application that supports DNSSEC in RHEL and Fedora blog post.
How it works
Note that we are assuming that you are using NetworkManager for your network configuration. For using or testing dnssec-trigger, we recommend using a fully updated Fedora 22 or later.
The solution we are using in Fedora consists of three components:
The purpose of each component is outlined in the following sections.
NetworkManager
In the client side DNSSEC deployment we use NetworkManager as the single source of the network configuration. We also rely on NetworkManager, to be notified on every network configuration change. This is really important, since the locally running validating resolver needs to be always configured properly to be able to successfully resolve domain names.
For communication with NetworkManager we use the Python bindings for libnm-glib. We are consuming the following information:
- current list of active connections
check the Connectivity state (if there is Captive Portal, etc.)not yet done- for each connection we inspect
- if the connection is marked as default by NetworkManager for IPv4 or IPv6 connectivity
- connection provided DNS resolvers
- connection provided (search) domains
- if the connection is VPN
dnssec-trigger
dnssec-trigger daemon and script are responsible for dynamic reconfiguration of Unbound server based on the current network configuration.
dnssec-trigger script
TBD
dnssec-trigger daemon
TBD
dnssec-trigger panel
TBD
Unbound
TBD
Installation
dnf install dnssec-trigger
This will get you all necessary packages (most importantly Unbound server) as dependencies.
Start using it
Since Fedora 22, dnssec-triggerd service is enabled by default and thus you can simply reboot. If that is inconvenient, start dnssec-triggerd service manually.
systemctl start dnssec-triggerd
Make sure dnssec-trigger-panel gets started in your session. It's an important piece of the ecosystem as it notifies you when DNSSEC trigger cannot be used and allows you to perform hotspot signon or disable DNSSEC temporarily. Having user interface at hand is a critical feature when moving between networks. On a headless system with an SSH session, the same actions can be performed using dnssec-trigger-control.
Get rid of it
The best way to disable the split DNS and DNSSEC functions temporarily is to choose Hotspot signon in the context menu of dnssec-trigger applet to which dnssec-trigger daemon currently responds by doing its best to reset /etc/resolv.conf to a state without dnssec-trigger. Stopping the daemon results in similar behavior. You can perform the same action from the command line as well.
dnssec-trigger-control hotspot_signon
To get back the above functions, you can use the same context menu, just choose Reprobe. Command line variant is also available.
dnssec-trigger-control reprobe
To abandon the features permanently, the best way is to uninstall dnssec-trigger and if you don't need unbound for other purposes, you can uninstall it as well.
yum remove unbound
Configuring unbound manually for split DNS and DNSSEC
Manual configuration via Unbound
TBD
Local zones
TBD
Global zone
TBD
Using dnssec-trigger-control (for testing only)
dnssec-trigger configures /etc/resolv.conf to use a local unbound instance on 127.0.0.1 and Unbound to use a secure global zone with nameservers submitted through dnssec-trigger-control or, if those aren't suitable, using public nameservers run by Fedora or the upstream project.
It also performs captive portal (hotspot) detection and temporarily changes /etc/resolv.conf to include the nameservers of the local network directly. That unfortunately breaks the local zones used with any network interfaces including those that have nothing to do with the captive portal connection.
NetworkManager integration
TBD
Debugging
Show global configuration and connection zones configuration in unbound:
$ unbound-control forward $ unbound-control list_forwards
To check NetworkManager's view of the configuration, use:
$ nmcli connection show active $ nmcli connection show active <id/uuid>
Documentation TODO
- Adding search domains.
- Common changes that people may wish to make such as the add_wifi_provided_zones
- When caches are flushed and what triggers that.
- How issues such as VPN provided name servers are handled.
- How to restart the services correctly.