wroclaw-git-backups/nix
1
0
Fork 0
mirror of https://github.com/NixOS/nix synced 2025-06-25 06:31:14 +02:00
Code Activity

focus on golden-path covering most scenarios

Browse source 
This should handle installation scenarios we can handle with
anything resembling confidence. Goal is approximating the existing
setup--not enforcing a best-practice...

Approaches (+ installer-handled, - manual) and configs each covers:

+ no change needed; /nix OK on boot volume:
  All pre-Catalina (regardless of T2 or FileVault use)

+ create new unencrypted volume:
  Catalina, pre-T2, no FileVault

+ create new encrypted-at-rest volume:
  Catalina, pre-T2, FileVault
  Catalina, T2, no FileVault

- require user to pre-create encrypted volume
  Catalina, T2, FileVault
This commit is contained in:
Travis A. Everett 2020-05-14 21:59:10 -05:00 • committed by Daiderd Jordan
parent 477d7c2d07
commit 2b0a81d92d
No known key found for this signature in database
GPG key ID: D02435D05B810C96
3 changed files with 331 additions and 115 deletions
Download patch file Download diff file Expand all files Collapse all files

350
doc/manual/installation/installing-binary.xml
View file

@ -6,16 +6,30 @@
<title>Installing a Binary Distribution</title>
<para>If you are using Linux or macOS, the easiest way to install Nix
is to run the following command:
<para>
If you are using Linux or macOS versions up to 10.14 (Mojave), the
easiest way to install Nix is to run the following command:
</para>
<screen>
$ sh &lt;(curl https://nixos.org/nix/install)
</screen>
As of Nix 2.1.0, the Nix installer will always default to creating a
single-user installation, however opting in to the multi-user
installation is highly recommended.
<para>
If you're using macOS 10.15 (Catalina) or newer, consult
<link linkend="sect-macos-installation">the macOS installation instructions</link>
before installing.
</para>
<para>
As of Nix 2.1.0, the Nix installer will always default to creating a
single-user installation, however opting in to the multi-user
installation is highly recommended.
<!-- TODO: this explains *neither* why the default version is
single-user, nor why we'd recommend multi-user over the default.
True prospective users don't have much basis for evaluating this.
What's it to me? Who should pick which? Why? What if I pick wrong?
-->
</para>
<section xml:id="sect-single-user-installation">
@ -36,7 +50,7 @@ run this under your usual user account, <emphasis>not</emphasis> as
root. The script will invoke <command>sudo</command> to create
<filename>/nix</filename> if it doesn’t already exist. If you don’t
have <command>sudo</command>, you should manually create
<command>/nix</command> first as root, e.g.:
<filename>/nix</filename> first as root, e.g.:
<screen>
$ mkdir /nix
@ -47,7 +61,7 @@ The install script will modify the first writable file from amongst
<filename>.bash_profile</filename>, <filename>.bash_login</filename>
and <filename>.profile</filename> to source
<filename>~/.nix-profile/etc/profile.d/nix.sh</filename>. You can set
the <command>NIX_INSTALLER_NO_MODIFY_PROFILE</command> environment
the <envar>NIX_INSTALLER_NO_MODIFY_PROFILE</envar> environment
variable before executing the install script to disable this
behaviour.
</para>
@ -81,12 +95,10 @@ $ rm -rf /nix
<para>
You can instruct the installer to perform a multi-user
installation on your system:
<screen>
sh &lt;(curl https://nixos.org/nix/install) --daemon
</screen>
</para>
<screen>sh &lt;(curl https://nixos.org/nix/install) --daemon</screen>
<para>
The multi-user installation of Nix will create build users between
the user IDs 30001 and 30032, and a group with the group ID 30000.
@ -136,82 +148,253 @@ sudo rm /Library/LaunchDaemons/org.nixos.nix-daemon.plist
</section>
<section xml:id="sect-apfs-volume-installation">
<title>APFS Volume Installation</title>
<section xml:id="sect-macos-installation">
<title>macOS Installation</title>
<para>
The root filesystem is read-only as of macOS 10.15 Catalina, all writable
paths were moved to a separate data volume. This means creating or writing
to <filename>/nix</filename> is not allowed. While changing the default prefix
would be possible, it's a very intrusive change that has side effects we want to
avoid for now.
Starting with macOS 10.15 (Catalina), the root filesystem is read-only.
This means <filename>/nix</filename> can no longer live on your system
volume, and that you'll need a workaround to install Nix.
</para>
<para>
For common writable locations <literal>firmlinks</literal> were introduced,
described by Apple as a "bi-directional wormhole" between two filesystems.
Essentially a bind mount for APFS volumes. However this is (currently) not
user configurable and only available for paths like <filename>/Users</filename>.
The recommended approach, which creates an unencrypted APFS volume
for your Nix store and a "synthetic" empty directory to mount it
over at <filename>/nix</filename>, is least likely to impair Nix
or your system.
</para>
<note><para>
With all separate-volume approaches, it's possible something on
your system (particularly daemons/services and restored apps) may
need access to your Nix store before the volume is mounted. Adding
additional encryption makes this more likely.
</para></note>
<para>
For special cases like NFS mount points or package manager roots <link xlink:href="https://developer.apple.com/library/archive/documentation/System/Conceptual/ManPages_iPhoneOS/man5/synthetic.conf.5.html">synthetic.conf(5)</link>
provides a mechanism for some limited, user-controlled file-creation at <filename>/</filename>.
This only applies at boot time, however <command>apfs.util</command> can be used
to trigger the creation (not deletion) of new entries without a reboot.
It would be ideal if this could create firmlinks, however a symlink or mountpoint
are the only options.
If you're using a recent Mac with a
<link xlink:href="https://www.apple.com/euro/mac/shared/docs/Apple_T2_Security_Chip_Overview.pdf">T2 chip</link>,
your drive will still be encrypted at rest (in which case "unencrypted"
is a bit of a misnomer). To use this approach, just install Nix with:
</para>
<screen>
alice$ /System/Library/Filesystems/apfs.fs/Contents/Resources/apfs.util -B
</screen>
<screen>$ sh &lt;(curl https://nixos.org/nix/install) --darwin-use-unencrypted-nix-store-volume</screen>
<itemizedlist>
<listitem>
<para>
The simplest solution is creating a symlink with <filename>/etc/synthetic.conf</filename>
to the data volume. (not recommended)
</para>
<para>
If you don't like the sound of this, you'll want to weigh the
other approaches and tradeoffs detailed in this section.
</para>
<screen>
nix /System/Volumes/Data/nix
</screen>
<note>
<title>Eventual solutions?</title>
<para>
All of the known workarounds have drawbacks, but we hope
better solutions will be available in the future. Some that
we have our eye on are:
</para>
<orderedlist>
<listitem>
<para>
A true firmlink would enable the Nix store to live on the
primary data volume without the build problems caused by
the symlink approach. End users cannot currently
create true firmlinks.
</para>
</listitem>
<listitem>
<para>
If the Nix store volume shared FileVault encryption
with the primary data volume (probably by using the same
volume group and role), FileVault encryption could be
easily supported by the installer without requiring
manual setup by each user.
</para>
</listitem>
</orderedlist>
</note>
<screen>
alice$ ls -l /
lrwxr-xr-x 1 root wheel 25 Jan 1 2019 nix -> /System/Volumes/Data/nix
</screen>
<section xml:id="sect-macos-installation-change-store-prefix">
<title>Change the Nix store path prefix</title>
<para>
Changing the default prefix for the Nix store is a simple
approach which enables you to leave it on your root volume,
where it can take full advantage of FileVault encryption if
enabled. Unfortunately, this approach also opts your device out
of some benefits that are enabled by using the same prefix
across systems:
<para>
However builds that detect or resolve this symlink will leak the canonical
location or even fail in certain cases, making this approach undesirable.
</para>
</listitem>
<itemizedlist>
<listitem>
<para>
Your system won't be able to take advantage of the binary
cache (unless someone is able to stand up and support
duplicate caching infrastructure), which means you'll
spend more time waiting for builds.
</para>
</listitem>
<listitem>
<para>
It's harder to build and deploy packages to Linux systems.
</para>
</listitem>
<!-- TODO: may be more here -->
</itemizedlist>
<listitem>
<para>
An empty directory can also be created using <filename>/etc/synthetic.conf</filename>,
this won't be writable but can be used as a mount point. And with
<literal>APFS</literal> it's relatively easy to create an separate
volume for nix instead.
</para>
<!-- TODO: Yes, but how?! -->
<screen>
nix
</screen>
It would also possible (and often requested) to just apply this
change ecosystem-wide, but it's an intrusive process that has
side effects we want to avoid for now.
<!-- magnificent hand-wavy gesture -->
</para>
<para>
</para>
</section>
<screen>
alice$ /System/Library/Filesystems/apfs.fs/Contents/Resources/apfs.util -B
alice$ sudo diskutil apfs addVolume diskX APFS 'Nix Store' -mountpoint /nix
alice$ mount
/dev/disk1s6 on /nix (apfs, local, journaled)
</screen>
<section xml:id="sect-macos-installation-encrypted-volume">
<title>Use a separate encrypted volume</title>
<para>
If you like, you can also add encryption to the recommended
approach taken by the installer. You can do this by pre-creating
an encrypted volume before you run the installer--or you can
run the installer and encrypt the volume it creates later.
<!-- TODO: see later note about whether this needs both add-encryption and from-scratch directions -->
</para>
<para>
In either case, adding encryption to a second volume isn't quite
as simple as enabling FileVault for your boot volume. Before you
dive in, there are a few things to weigh:
</para>
<orderedlist>
<listitem>
<para>
The additional volume won't be encrypted with your existing
FileVault key, so you'll need another mechanism to decrypt
the volume.
</para>
</listitem>
<listitem>
<para>
You can store the password in Keychain to automatically
decrypt the volume on boot--but it'll have to wait on Keychain
and may not mount before your GUI apps restore. If any of
your launchd agents or apps depend on Nix-installed software
(for example, if you use a Nix-installed login shell), the
restore may fail or break.
</para>
<para>
On a case-by-case basis, you may be able to work around this
problem by using <command>wait4path</command> to block
execution until your executable is available.
</para>
<para>
It's also possible to decrypt and mount the volume earlier
with a login hook--but this mechanism appears to be
deprecated and its future is unclear.
</para>
</listitem>
<listitem>
<para>
You can hard-code the password in the clear, so that your
store volume can be decrypted before Keychain is available.
</para>
</listitem>
</orderedlist>
<para>
If you are comfortable navigating these tradeoffs, you can encrypt the volume with
something along the lines of:
<!-- TODO:
I don't know if this also needs from-scratch instructions?
can we just recommend use-the-installer-and-then-encrypt?
-->
</para>
<!--
TODO: it looks like this option can be encryptVolume|encrypt|enableFileVault
<para>
This does make the installation more complicated, requiring both
<filename>/etc/synthetic.conf</filename> as well as <filename>/etc/fstab</filename>
</para>
It may be more clear to use encryptVolume, here? FileVault seems
heavily associated with the boot-volume behavior; I worry
a little that it can mislead here, especially as it gets
copied around minus doc context...?
-->
<screen>alice$ diskutil apfs enableFileVault /nix -user disk</screen>
<!-- TODO: and then go into detail on the mount/decrypt approaches? -->
</section>
<section xml:id="sect-macos-installation-symlink">
<!--
Maybe a good razor is: if we'd hate having to support someone who
installed Nix this way, it shouldn't even be detailed?
-->
<title>Symlink the Nix store to a custom location</title>
<para>
Another simple approach is using <filename>/etc/synthetic.conf</filename>
to symlink the Nix store to the data volume. This option also
enables your store to share any configured FileVault encryption.
Unfortunately, builds that resolve the symlink may leak the
canonical path or even fail.
</para>
<para>
Because of these downsides, we can't recommend this approach.
</para>
<!-- Leaving out instructions for this one. -->
</section>
<section xml:id="sect-macos-installation-recommended-notes">
<title>Notes on the recommended approach</title>
<para>
This section goes into a little more detail on the recommended
approach. You don't need to understand it to run the installer,
but it can serve as a helpful reference if you run into trouble.
</para>
<orderedlist>
<listitem>
<para>
In order to compose user-writable locations into the new
read-only system root, Apple introduced a new concept called
<literal>firmlinks</literal>, which it describes as a
"bi-directional wormhole" between two filesystems. You can
see the current firmlinks in <filename>/usr/share/firmlinks</filename>.
Unfortunately, firmlinks aren't (currently?) user-configurable.
</para>
<para>
For special cases like NFS mount points or package manager roots,
<link xlink:href="https://developer.apple.com/library/archive/documentation/System/Conceptual/ManPages_iPhoneOS/man5/synthetic.conf.5.html">synthetic.conf(5)</link>
supports limited user-controlled file-creation (of symlinks,
and synthetic empty directories) at <filename>/</filename>.
To create a synthetic empty directory for mounting at <filename>/nix</filename>,
add the following line to <filename>/etc/synthetic.conf</filename>
(create it if necessary):
</para>
<screen>nix</screen>
</listitem>
<listitem>
<para>
This configuration is applied at boot time, but you can use
<command>apfs.util</command> to trigger creation (not deletion)
of new entries without a reboot:
</para>
<screen>alice$ /System/Library/Filesystems/apfs.fs/Contents/Resources/apfs.util -B</screen>
</listitem>
<listitem>
<para>
Create the new APFS volume with diskutil:
</para>
<screen>alice$ sudo diskutil apfs addVolume diskX APFS 'Nix Store' -mountpoint /nix</screen>
</listitem>
<listitem>
<para>
Using <command>vifs</command>, add the new mount to
<filename>/etc/fstab</filename>. If it doesn't already have
other entries, it should look something like:
</para>
<screen>
#
@ -219,29 +402,16 @@ alice$ mount
#
# Failure to do so is unsupported and may be destructive.
#
LABEL=Nix\040Store /nix apfs rw
LABEL=Nix\040Store /nix apfs rw,nobrowse
</screen>
<para>
On macOS volumes are also mounted quite late, launchd services or other
things that start during login will start before our volume is mounted.
For these cases eg. <command>wait4path</command> must be used for
things that depend on <filename>/nix</filename>.
</para>
<para>
This new volume also won't be encrypted by default, and enabling it
requires extra setup. For machines with a <link xlink:href="https://www.apple.com/euro/mac/shared/docs/Apple_T2_Security_Chip_Overview.pdf">T2 chip</link>
all data is already entrypted at rest, older hardware won't even when
FileVault is enabled for the rest of the system.
</para>
<screen>
alice$ diskutil apfs enableFileVault /nix -user disk
</screen>
</listitem>
</itemizedlist>
<para>
The nobrowse setting will keep Spotlight from indexing this
volume, and keep it from showing up on your desktop.
</para>
</listitem>
</orderedlist>
</section>
</section>