title: “Nix Installation on macOS Catalina” date: 2020-06-14T12:39:29+03:00 draft: true
This is kind of a part 1.5 to Getting Started with Nix, we’re going into a bit more of details regarding installation of Nix on macOS. This is also an execuse to dig in a little and write about Apple’s changes to the filesystem in macOS Catalina, and other APFS concepts.
This can totally be skipped if you’re not on any Linux based system, or if you don’t care much about how APFS handles things under the hood and would rather to just blindly follow the instructions given by the Nix installer!
APFS is Apple’s filesystem on iOS (since iOS 10.3) and macOS (since 10.13 High Sierra). Conceptually APFS divides your disc into 2 layers, a container layer and a file system (volumes) layer.
The container layer exists within the disk’s partitioning scheme and is the more high level of the two, it provides checkpoints for the actions executed in the drive, ensures those actions are atomic, and provides crash protection, among other things. It also contains volume metadata and snapshots, as well its encryption state.
The file-system layer, or the volume, contains the actual data structures that store information, you know your directory structures, file metadata, and file content. In APFS, Volumes can be assigned specific roles, one of 16 roles to be exact (as of macOS Catalina 10.15). This allows the system to treat each volume in a more optimized manner depending on its role, examples on the roles are preboot, virtual memory, or recovery volumes, those three were one of the original volume roles in APFS.
Partitions in APFS contain exactly one container, and containers contain inside them the volumes (file system layers) that host your actual data. Containers allow the volumes inside them to “share” the space allocated to the container, and in a sense allows each volume to dynamically expand or shrink as needed, within the confides of the container itself. However containers themselves requires more manual labor and are more risky to resize, as they can only be expanded in one direction and only into adjacent free space.
As an example of the quality of life improvements introduced by containers allowing inner volumes to resize dynamic, taken from Apples APFS reference, a drive can be configured with two bootable volumes — one with a shipping version of macOS and one with a beta version — as well as a user data volume. All three of these volumes share free space, meaning you donʼt have to decide ahead of time how to divide space between them.
After the release of macOS Catalina, Apple introduces a fancy new concept to APFS called volume groups. Volume groups are a logical grouping of multiple volumes within a container, it allows the system to treat them in a special way, and creates the illusion that these two separate volumes are one.
Following is a diagram of APFS, courtesy of https://bombich.com
Volume groups are directly observable when you upgrade to macOS Catlina, as you will notice if you open the Disk Utility that the volume your OS used to have is split in two, “Macintosh HD” and “Macintosh HD - Data”. Even though from a user precpetive they are treated as one single entity, both in Finder and in the terminal.
As means to provide an extra layer of security, the OS itself is installed in a volume with the system read only role,
APFS_VOL_ROLE_SYSTEM. This makes it practically impossible for an attacker to overwrite or manipulate the OS installation!
Not everything can reside in a volume that is readonly, that’s why there exists a nother volume that contains all of the user data, as well as any system portions that require write access. This volume is always mounted but is effectively hidden from the users, in Finder it is practically mashed together with the system volume as to appear as one volume.
In order to make the volumes split transparent APFS supports what’s called firmlinks. A firmlink conceptually similar symlink, except it only works on folders. It creates a “bi-directional wormwhole” that transparently merges folders across volumes and handle, transparently, the transition between one volume to the other when moving in the directory structure!
An great example of this is the
/Users folder, it appears to reside in the root of the filesystem, but in reality it exists in the data volume, the
/Users we see is a firmlink between the system readonly volume and the actual folder in the data volume!
You can list all system firmlinks by viewing the
/usr/share/firmlinks file. In this diargram taken from the Apple WWDC2019 presentation on APFS changes, you can see an example of how firmlinks are placed.
Alright, alright, that’s like cool I guess or whatever, but how do I write to the root of my system? How do I add a new folder under
/? Welp, you kinda can’t, or rather you can but you have to go through something called
You see, since the root volume is now readonly, you literally cannot write to it, unless you go jumping through hoops to reenable writing to it by disabling SIP, rebooting, re
mounting to volume with write access, doing the write stuff. Oh, and none of that persists between reboots, so … yaay! Let’s just ignore that this is even an option, and focus on the next best thing,
So remember firmlinks from before right, macOS provides a mechanism for creating user controlled firmlinks. Those “synthetic links” can be created by adding entries to the
/etc/synthetic.conf. Those entries follow the same formate found in the system firmlinks in
/usr/share/firmlinks, except those are meant to be managed by the user for whatever purposes there are!
Alright let’s do just that, let’s create a folder in the system volume were we can write stuff freely! It’s relatively simple to do so, first we need to add an entry to the
/etc/synthetic.conf file declaring the folder we wanna create, say
/blah, now the syntax of synthetic.conf is simple, it’s a tab separated list of files, the following is the syntax of this file, for more details you can open a terminal and run
man synthetic.conf it’s a very short on point manual.
Ok, so we can just run this command
echo 'blah Users/<your-user-name>/blah' | sudo tee -a /etc/synthetic.conf and magically we’ll create this file/filder under root if missing and link it up with the writable folder under our user! One reboot later, and baam! We have ourselves a writable folder under
We can now write to root, will not really but close enough for pretty much all pruposes!
# if we list folder in the root and grep for blah here's our synthetic link $ ls / | grep blah lrwxr-xr-x 1 root wheel 14 Jun 20 23:20 blah -> Users/oji/blah # trying to create a file under it fails since the actual folder doesn't exist $ touch /blah/foobarr touch: /blah/foobarr: No such file or directory $ mkdir ~/blah $ touch /blah/foobarr $ ls /blah lrwxr-xr-x 1 root wheel 14 Jun 20 23:20 /blah -> Users/oji/blah $ ls /blah/ total 0 -rw-r--r-- 1 oji staff 0 Jun 27 16:10 foobarr $ ls ~/blah/ total 0 -rw-r--r-- 1 oji staff 0 Jun 27 16:10 foobarr
Things to note here are the strict use of tabs in the synthetic.conf file, and the fact that files aren’t preceeded by
/ in their entries, adding
/ would cause them to not be created. I assume this is because they are always assumed to be relative to the root folder.
Also notice how these user created firmlinks are behaving like symlink here. In truth I kinda lied when I called them user created firmlink, firmlinks are exclusively reserved only for the OS those are “synthetic symbolic link” they are virtual symlinks that are managed by the OS as if they are firmlinks, but they behave more similarly to normal symlinks to the end users, aside from the fact they can only be created/destroyed at boot time.
Ok, what the hell does any of that have to do with Nix? Well Nix uses the folder
/nix for it’s installation, as well as everything else it ever writes to disk. As you can tell this means that Nix needs write access to the now read-only system volume.
The officially endorsed way for installing Nix on macOS is by utilizing the
synthetic.conf to add a firmlink under
/nix that points to a user writable volume. This can be done manually if you want, or automatically by utilizing the