Impermanence

• Contact
• System setup
  • tmpfs
  • BTRFS subvolumes
• Module usage
  • NixOS
  • home-manager
  • Further reading

Lets you choose what files and directories you want to keep between reboots - the rest are thrown away.

Why would you want this?

• It keeps your system clean by default.
• It forces you to declare settings you want to keep.
• It lets you experiment with new software without cluttering up your system.

There are a few different things to set up for this to work:

• A root filesystem which somehow gets wiped on reboot. There are a few ways to achieve this. See the System setup section for more info.
• At least one mounted volume where the files and directories you want to keep are stored permanently.
• At least one of the modules in this repository, which take care of linking or bind mounting files between the persistent storage mount point and the root file system. See the Module usage section for more info.

Contact

Join the matrix room to chat about the project.

System setup

There are many ways to wipe your root partition between boots. This section lists a few common ways to accomplish this, but is by no means an exhaustive list.

tmpfs

The easiest method is to use a tmpfs filesystem for the root. This is the easiest way to set up impermanence on systems which currently use a traditional filesystem (ext4, xfs, etc) as the root filesystem, since you don't have to repartition.

All data stored in tmpfs only resides in system memory, not on disk. This automatically takes care of cleaning up between boots, but also comes with some pretty significant drawbacks:

• Downloading big files or trying programs that generate large amounts of data can easily result in either an out-of-memory or disk-full scenario.
• If the system crashes or loses power before you've had a chance to move files you want to keep to persistent storage, they're gone forever.

Using tmpfs as the root filesystem, the filesystem setup would look something like this:

  {
    fileSystems."/" = {
      device = "none";
      fsType = "tmpfs";
      options = [ "defaults" "size=25%" "mode=755" ];
    };

    fileSystems."/persistent" = {
      device = "/dev/root_vg/root";
      neededForBoot = true;
      fsType = "btrfs";
      options = [ "subvol=persistent" ];
    };

    fileSystems."/nix" = {
      device = "/dev/root_vg/root";
      fsType = "btrfs";
      options = [ "subvol=nix" ];
    };

    fileSystems."/boot" = {
      device = "/dev/disk/by-uuid/XXXX-XXXX";
      fsType = "vfat";
    };
  }

where the size option determines how much system memory is allowed to be used by the filesystem.

BTRFS subvolumes

A more advanced solution which doesn't have the same drawbacks as using tmpfs is to use a regular filesystem, but clean it up between boots. A relatively easy way to do this is to use BTRFS and create a new subvolume to use as root on boot. This also allows you to keep a number of old roots around, in case of crashes, power outages or other accidents.

A setup which would remove automatically remove roots that are older than 30 days could look like this:

  {
    fileSystems."/" = {
      device = "/dev/root_vg/root";
      fsType = "btrfs";
      options = [ "subvol=root" ];
    };

    boot.initrd.postDeviceCommands = lib.mkAfter ''
      mkdir /btrfs_tmp
      mount /dev/root_vg/root /btrfs_tmp
      if [[ -e /btrfs_tmp/root ]]; then
          mkdir -p /btrfs_tmp/old_roots
          timestamp=$(date --date="@$(stat -c %Y /btrfs_tmp/root)" "+%Y-%m-%-d_%H:%M:%S")
          mv /btrfs_tmp/root "/btrfs_tmp/old_roots/$timestamp"
      fi

      delete_subvolume_recursively() {
          IFS=$'\n'
          for i in $(btrfs subvolume list -o "$1" | cut -f 9- -d ' '); do
              delete_subvolume_recursively "/btrfs_tmp/$i"
          done
          btrfs subvolume delete "$1"
      }

      for i in $(find /btrfs_tmp/old_roots/ -maxdepth 1 -mtime +30); do
          delete_subvolume_recursively "$i"
      done

      btrfs subvolume create /btrfs_tmp/root
      umount /btrfs_tmp
    '';

    fileSystems."/persistent" = {
      device = "/dev/root_vg/root";
      neededForBoot = true;
      fsType = "btrfs";
      options = [ "subvol=persistent" ];
    };

    fileSystems."/nix" = {
      device = "/dev/root_vg/root";
      fsType = "btrfs";
      options = [ "subvol=nix" ];
    };

    fileSystems."/boot" = {
      device = "/dev/disk/by-uuid/XXXX-XXXX";
      fsType = "vfat";
    };
  }

This assumes the BTRFS filesystem can be found in an LVM volume group called root_vg. Adjust the path as necessary.

Module usage

There are currently two modules: one for NixOS and one for home-manager.

NixOS

To use the module, import it into your configuration with

  {
    imports = [ /path/to/impermanence/nixos.nix ];
  }

or use the provided nixosModules.impermanence flake output:

  {
    inputs = {
      impermanence.url = "github:nix-community/impermanence";
    };

    outputs = { self, nixpkgs, impermanence, ... }:
      {
        nixosConfigurations.sythe = nixpkgs.lib.nixosSystem {
          system = "x86_64-linux";
          modules = [
            impermanence.nixosModules.impermanence
            ./machines/sythe/configuration.nix
          ];
        };
      };
  }

This adds the environment.persistence option, which is an attribute set of submodules, where the attribute name is the path to persistent storage.

Usage is shown best with an example:

  {
    environment.persistence."/persistent" = {
      enable = true;  # NB: Defaults to true, not needed
      hideMounts = true;
      directories = [
        "/var/log"
        "/var/lib/bluetooth"
        "/var/lib/nixos"
        "/var/lib/systemd/coredump"
        "/etc/NetworkManager/system-connections"
        { directory = "/var/lib/colord"; user = "colord"; group = "colord"; mode = "u=rwx,g=rx,o="; }
      ];
      files = [
        "/etc/machine-id"
        { file = "/var/keys/secret_file"; parentDirectory = { mode = "u=rwx,g=,o="; }; }
      ];
      users.talyz = {
        directories = [
          "Downloads"
          "Music"
          "Pictures"
          "Documents"
          "Videos"
          "VirtualBox VMs"
          { directory = ".gnupg"; mode = "0700"; }
          { directory = ".ssh"; mode = "0700"; }
          { directory = ".nixops"; mode = "0700"; }
          { directory = ".local/share/keyrings"; mode = "0700"; }
          ".local/share/direnv"
        ];
        files = [
          ".screenrc"
        ];
      };
    };
  }

• "/persistent" is the path to your persistent storage location This allows for multiple different persistent storage locations. If you, for example, have one location you back up and one you don't, you can use both by defining two separate attributes under environment.persistence.
• enable determines whether the persistent storage location should be enabled or not. Useful when sharing configurations between systems with and without impermanence setups. Defaults to true.
• hideMounts allows you to specify whether to hide the bind mounts from showing up as mounted drives in the file manager. If enabled, it sets the mount option x-gvfs-hide on all the bind mounts.

• directories are all directories you want to bind mount to persistent storage. A directory can be represented either as a string, simply denoting its path, or as a submodule. The submodule representation is useful when the default assumptions, mainly regarding permissions, are incorrect. The available options are:

  • directory, the path to the directory you want to bind mount to persistent storage. Only setting this option is equivalent to the string representation.
  • persistentStoragePath, the path to persistent storage. Defaults to the environment.persistence submodule name, i.e. "/persistent" in the example. This should most likely be left to its default value - don't change it unless you're certain you really need to.
  • user, the user who should own the directory. If the directory doesn't already exist in persistent storage, it will be created and this user will be its owner. This also applies to any parent directories which don't yet exist. Changing this once the directory has been created has no effect.
  • group, the group who should own the directory. If the directory doesn't already exist in persistent storage, it will be created and this group will be its owner. This also applies to any parent directories which don't yet exist. Changing this once the directory has been created has no effect.
  • mode, the permissions to set for the directory. If the directory doesn't already exist in persistent storage, it will be created with this mode. Can be either an octal mode (e.g. 0700) or a symbolic mode (e.g. u=rwx,g=,o=). Parent directories that don't yet exist are created with default permissions. Changing this once the directory has been created has no effect.

• files are all files you want to link or bind to persistent storage. A file can be represented either as a string, simply denoting its path, or as a submodule. The submodule representation is useful when the default assumptions, mainly regarding the permissions of its parent directory, are incorrect. The available options are:

  • file, the path to the file you want to bind mount to persistent storage. Only setting this option is equivalent to the string representation.
  • persistentStoragePath, the path to persistent storage. Defaults to the environment.persistence submodule name, i.e. "/persistent" in the example. This should most likely be left to its default value - don't change it unless you're certain you really need to.
  • parentDirectory, the permissions that should be applied to the file's parent directory, if it doesn't already exist. Available options are user, group and mode. See their definition in directories above.

  If the file exists in persistent storage, it will be bind mounted to the target path; otherwise it will be symlinked.

• users.talyz handles files and directories in talyz's home directory The users option defines a set of submodules which correspond to the users' names. The directories and files options of each submodule work like their root counterparts, but the paths are automatically prefixed with with the user's home directory. If the user has a non-standard home directory (i.e. not /home/<username>), the users.<username>.home option has to be set to this path - it can't currently be automatically deduced due to a limitation in nixpkgs.

Important note: Make sure your persistent volumes are marked with neededForBoot, otherwise you will run into problems.

home-manager

Usage of the home-manager module is very similar to the one of the NixOS module - the key differences are that the persistence option is now under home, rather than environment, and the addition of the submodule option removePrefixDirectory.

Important note: You have to use the home-manager NixOS module (in the nixos directory of home-manager's repo) in order for this module to work as intended.

To use the module, import it into your configuration with

  {
    imports = [ /path/to/impermanence/home-manager.nix ];
  }

This adds the home.persistence option, which is an attribute set of submodules, where the attribute name is the path to persistent storage.

Usage is shown best with an example:

  {
    home.persistence."/persistent/home/talyz" = {
      directories = [
        "Downloads"
        "Music"
        "Pictures"
        "Documents"
        "Videos"
        "VirtualBox VMs"
        ".gnupg"
        ".ssh"
        ".nixops"
        ".local/share/keyrings"
        ".local/share/direnv"
        {
          directory = ".local/share/Steam";
          method = "symlink";
        }
      ];
      files = [
        ".screenrc"
      ];
      allowOther = true;
    };
  }

• "/persistent/home/talyz" is the path to your persistent storage location

• directories are all directories you want to link to persistent storage

  • It is possible to switch the linking method between bindfs (the default) and symbolic links.
• files are all files you want to link to persistent storage. These are symbolic links to their target location.
• allowOther allows other users, such as root, to access files through the bind mounted directories listed in directories. Useful for sudo operations, Docker, etc. Requires the NixOS configuration programs.fuse.userAllowOther = true.

Additionally, the home-manager module allows for compatibility with dotfiles repos structured for use with GNU Stow, where the files linked to are one level deeper than where they should end up. This can be achieved by setting removePrefixDirectory to true:

  {
    home.persistence."/etc/nixos/home-talyz-nixpkgs/dotfiles" = {
      removePrefixDirectory = true;
      files = [
        "screen/.screenrc"
      ];
      directories = [
        "fish/.config/fish"
      ];
    };
  }

In the example, the .screenrc file and .config/fish directory should be linked to from the home directory; removePrefixDirectory removes the first part of the path when deciding where to put the links.

Note: When using bindfs fuse filesystem for directories, the names of the directories you add will be visible in the /etc/mtab file and in the output of mount to all users.

Further reading

The following blog posts provide more information on the concept of ephemeral roots:

• https://elis.nu/blog/2020/05/nixos-tmpfs-as-root/ — @etu's blog post walks the reader through a NixOS-on-tmpfs installation.
• https://grahamc.com/blog/erase-your-darlings — @grahamc's blog post details why one would want to erase their state at every boot, as well as how to achieve this using ZFS snapshots.
• https://willbush.dev/blog/impermanent-nixos/ — @willbush's blog post provides a detailed NixOS-on-tmpfs guide with optional LUKS encryption, and utilizing nix flakes for an opinionated install.