Previously we just applied a very minimal set of restrictions and
trusted unbound to properly drop root privs and capabilities.
With this change I am (for the most part) just using the upstream
example unit file for unbound. The main difference is that we start
unbound was `unbound` user with the required capabilities instead of
letting unbound do the chroot & uid/gid changes.
The upstream unit configuration this is based on is a lot stricter with
all kinds of permissions then our previous variant. It also came with
the default of having the `Type` set to `notify`, therefore we are also
using the `unbound-with-systemd` package here. Unbound will start up,
read the configuration files and start listening on the configured ports
before systemd will declare the unit "running". This will likely help
with startup order and the occasional race condition during system
activation where the DNS service is started but not yet ready to answer
queries.
Aditionally to the much stricter runtime environmet I removed the
`/dev/urandom` mount lines we previously had in the code (that would
randomly fail during `stop`-phase).
The `preStart` script is now only required if we enabled the trust
anchor updates (which are still enabled by default).
Another beneefit of the refactoring is that we can now issue reloads via
either `pkill -HUP unbound` or `systemctl reload unbound` to reload the
running configuration without taking the daemon offline. A prerequisite
of this was that unbound configuration is available on a well known path
on the file system. I went for /etc/unbound/unbound.conf as that is the
default in the CLI tooling which in turn enables us to use
`unbound-control` without passing a custom configuration location.
This is a refactor of how resolvconf is managed on NixOS. We split it
into a separate service which is enabled internally depending on whether
we want /etc/resolv.conf to be managed by it. Various services now take
advantage of those configuration options.
We also now use systemd instead of activation scripts to update
resolv.conf.
NetworkManager now uses the right option for rc-manager DNS
automatically, so the configuration option shouldn't be exposed.
Rather than special-casing the dns options in networkmanager.nix, use
the module system to let unbound and systemd-resolved contribute to
the newtorkmanager config.
While entering the chroot should provide the same amount of isolation,
the preStart script will run with full root privileges and so would
benefit from some isolation as well (in particular due to
unbound-anchor, which can perform network I/O).
1. The preStart script ensures consistent ownership, even if the unbound
user's uid has changed
2. The unbound daemon does not generate data that needs to be private to
it, so it would not matter that a different service would end up
owning its data (as long as unbound remains enabled, it should reclaim
ownership soon enough anyway).
Thus, there's no clear benefit to allocate a dedicated uid for the
unbound service. This releases uid/gid 48.
Also, because the preStart script creates the data directory, there's no
need to specify a homedir or ask for its creation.
/dev/random is an exhaustible resource. Presumably, unbound will not be
used to generate long-term encryption keys and so allowing it to use
/dev/random only increases the risk of entropy exhaustion for no
benefit.
Using pkgs.lib on the spine of module evaluation is problematic
because the pkgs argument depends on the result of module
evaluation. To prevent an infinite recursion, pkgs and some of the
modules are evaluated twice, which is inefficient. Using ‘with lib’
prevents this problem.
