feat: Impermanence (system)

I had to disable initrd's systemd stuff.  I just /couldn't/ get a rollback
service working; believe me, not for a lack of effort!  I've been working on
this on-and-off for a month or two now.
This commit is contained in:
Madeleine Sydney
2025-01-01 15:45:10 -07:00
parent dfa5d6625b
commit 9516c35c7f
7 changed files with 109 additions and 381 deletions

View File

@@ -11,10 +11,6 @@ A second try at NixOS, now that I have a better idea of what I'm doing. The effo
* Inbox
** Refactor: single file per feature
Modules, but lightweight.
* Principles
** User configuration
@@ -75,7 +71,17 @@ Where =default.nix= returns an attrset of form
** Impermanence and persistence
- Persistent files to be linked into ~/~ go under ~/persist/root~
I use impermanence to wipe most of my filesystem on boot.
*** Boot process
What follows is an overview of [[file:modules/nixos/impermanence/rollback.nix][modules/nixos/impermanence/rollback.nix]].
On boot, ...
- The existing subvolume root filesystem will be moved to a 'death row' directory, where it will live for about three days before deletion. Precisely, =«btrfs-filesystem»/«root-subvolume»= is moved to =«btrfs-filesystem»/old-roots/«timestamp»=. The brief grace period allows for easy recovery in the (very common) case where files are unintentionally deleted due to the user's silly human negligence.
- A new, blank subvolume is created in place of the previous. Precisely, the subvolume =«btrfs-filesystem»/«root-subvolume»= is created.
- Any subvolumes under =«btrfs-filesystem»/old-roots= older than three days are deleted.
* Tasks
@@ -301,6 +307,8 @@ https://discourse.nixos.org/t/what-to-do-with-a-full-boot-partition/2049
- If neither are available, use ~nix run nixpkgs#nvim~
- If this fails, try ~nano~.
- Support ~--pager~
*** TODO =rage-edit=
*** TODO ~forget-host HOST~

View File

@@ -17,7 +17,7 @@
];
impermanence = {
enable = false;
enable = true;
directories = [
# "Warning: Neither /var/lib/nixos nor any of its parents are persisted.
# This means all users/groups without specified uids/gids will have them
@@ -28,25 +28,15 @@
# just think it's annoying to edit ~/.ssh/known_hosts all the time.
"/etc/ssh"
];
device = "placeholderrrr";
archiveLimit = 3;
rollback = {
enable = true;
device = "/dev/sda2";
subvolume = "rootfs";
};
};
# niri.enable = true;
};
boot = {
initrd = {
enable = true;
systemd.enable = true;
# systemd.initrdBin = with pkgs; [
# zfs
# coreutils
# babashka
# ];
};
loader = {
systemd-boot.enable = true;
efi.canTouchEfiVariables = false;
@@ -60,28 +50,6 @@
mountPoint = "/persist/dots";
};
# HACK: This is here temporarily so I can use `/persist` without
# Impermanence. This can be removed once impermanence is enabled.
# Create a group called `cfg.persistGroupName`
users.groups.persist = {
name = "persist";
};
# HACK: This is here temporarily so I can use `/persist` without
# Impermanence. This can be removed once impermanence is enabled.
systemd.tmpfiles.settings = {
"10-persist" = {
# Permit members of `cfg.persistGroupName` to read, write, and execute
# /persist.
"/persist" = {
z = {
group = "persist";
mode = "2775";
};
};
};
};
# networking.hostName = "nixos-testbed";
networking.hostId = "238e9b1e"; # head -c 8 /etc/machine-id

View File

@@ -6,6 +6,9 @@
boot.initrd.supportedFilesystems.btrfs = true;
boot.supportedFilesystems.btrfs = true;
# From Impermanence's README: "Important note: Make sure your persistent
# volumes are marked with neededForBoot, otherwise you will run into
# problems."
fileSystems."/persist".neededForBoot = true;
disko.devices = {

View File

@@ -1,167 +0,0 @@
#!/usr/bin/env bb
;; TODO: rewrite with fewer assumptions about the filesystem structure. Perhaps
;; we can achieve near-atomicity by doing to bulk of the work with a function
;; `erase-home-darlings : FileSystem -> Maybe FileSystem`, which will not do any
;; IO.
;; TODO: option to either move OR copy
(require '[clojure.core.match :refer [match]]
'[babashka.cli :as cli]
'[clojure.pprint :as pp]
'[clojure.tools.logging :as l]
'[babashka.process :refer [shell check process] :as p])
(defn get-files [{:keys [rollback-to dataset]}]
(let [snapshot (str dataset "@" rollback-to)
diff (:out (shell {:out :string}
"zfs diff -HF"
snapshot
dataset))]
;; See zfs-diff(8) to understand what we're parsing here.
(->> diff
str/split-lines
(map #(str/split % #"\s"))
(filter #(and
;; We only care to preserve /new/ content.
(contains? #{"+" "M"} (first %))
;; We only bother with plain old files. No directories,
;; symlinks, etc.
(= (second %) "F")))
(map #(nth % 2)))))
(defn move-out-of-my-way [file]
;; No TCO. }:<
(let [maximum-attempts 50]
(loop [n 0]
(let [file' (format "%s-%d" file n)]
(if (fs/exists? file')
(do (printf (str "Failed to rename `%s' to `%s', "
"because the latter already exists.\n")
file file')
(if (< n maximum-attempts)
(recur (inc n))
(do (printf (str "We've tried to rename `%s' %d "
"without success. This should "
"never happen! Abort!")
file (inc n))
(System/exit 127))))
(do (fs/move file file')
file'))))))
(defn archive-files [{:keys [archive-to]} files]
(let [new-archive (fs/path archive-to "new-archive")]
(when (fs/exists? new-archive)
(println "Warning: `new-archive' already exists... we'll rename it for you?")
(move-out-of-my-way new-archive))
(doseq [file files]
(let [destination (fs/path
new-archive
(fs/relativize "/home" (fs/parent file)))]
(fs/create-dirs destination)
(fs/move file destination)))))
;; FIXME: This code could be a lot easier on the eyes. }:\
(defn cycle-archives [{:keys [archive-to archive-limit]}]
"Delete the oldest archive path, and increment each previous path by one.
More precisely,
- Delete the archive path labeled `n` (the oldest allowed).
- For each remaining path labeled 'i', relabel to 'i + 1'.
- Lastly, we delete the path labeled `new-archive`, if it exists."
(let [gp (memoize #(fs/path archive-to (str %)))]
(when (fs/exists? (gp archive-limit))
(fs/delete-tree (gp archive-limit)))
(doseq [i (range (dec archive-limit) 0 -1)]
(when (fs/exists? (gp i))
(fs/move (gp i) (gp (inc i)))))
(when (fs/exists? (gp "new-archive"))
(fs/move (gp "new-archive") (gp 1)))))
(defn do-rollback [{:keys [dataset rollback-to]}]
(let [proc (shell "zfs" "rollback" "-r" (str dataset "@" rollback-to))]
(if (= (:exit proc) 0)
(println (str "Successfully rolled back /home. "
"Enjoy that fresh filesystem smell! }:D"))
(println "Something went wrong rolling back /home... D:{"))))
(def zfs-dataset?
;; We memoise an anonymous procedure because we want the command to be run on
;; the first invocation of `zfs-dataset?`, *not* when this file's top-level is
;; first evaluated. Naïvely, it's reasonable to think we should instead do
;; something like
;; (def zfs-dataset?
;; (let [datasets <call zfs list>]
;; (fn [x] (contains? datasets x))))
;; but that would call zfs (and potentially throw) far too early to make sense.
(let [get-datasets (memoize
(fn []
(->> (:out (shell {:out :string} "ls" "-la"))
str/split-lines
(map #(first (str/split % #"\s")))
set)))]
(fn [x]
(contains? (get-datasets) x))))
(def cli-spec
{:spec
{:archive-limit {:coerce :int
:alias :n
:validate #(pos? %)
:default 3
:desc "Number of archives to save at a time."}
:dataset {:coerce :string
;; :validate zfs-dataset?
:require true
:desc "Dataset to be archived and rolled back."}
:rollback-to {:coerce :string
;; TODO: Validate snapshot.
:require true
:desc "Snapshot to rollback to."}
:archive-to {:coerce :string
:default "/persist/previous/home"
:desc "The path under which archives will be stored."}
:error-fn
(fn [{:keys [spec type cause msg option] :as data}]
(when (= :org.babashka/cli type)
(case cause
:require
(println
(format "Missing required argument: %s\n" option))))
(System/exit 1))}})
(defmacro with-echoed-shell-commands [& body]
(let [print-cmd #(println (str "+ " (str/join (:cmd %))))]
`(binding [p/*defaults* {:pre-start-fn ~print-cmd}]
~@body)))
(defmacro with-echoed-shell-commands [& body]
`(do ~@body))
(defn -main [opts]
(pp/pprint opts)
(with-echoed-shell-commands
(shell "mount" (:dataset opts) "/home")
(let [files (get-files opts)]
(archive-files opts files)
(cycle-archives opts)
(do-rollback opts))))
#_
(def fs-ops
{:zfs {:get-files zfs-get-files
:rollback zfs-rollback}})
#_
(defn -main [opts]
(let [test-bin (fn [x]
(printf "%s: %s\n"
x (map str (fs/which-all x))))]
(test-bin "mount")
(test-bin "findmnt")
(test-bin "zfs"))
(shell "mount")
(shell "ls -la /home"))
(-main (cli/parse-opts *command-line-args* cli-spec))

View File

@@ -1,4 +1,3 @@
# Requires boot.initrd.enable = true and boot.initrd.systemd.enable = true!
{ config, lib, pkgs, ... }:
with lib;
@@ -6,41 +5,38 @@ with lib;
let
cfg = config.sydnix.impermanence;
in {
imports = [
./impermanence/rollback.nix
];
options = {
sydnix.impermanence = {
enable = mkOption {
description = "Enable Impermanence";
type = types.bool;
default = false;
};
enable = mkEnableOption "Impermanence";
directories = mkOption {
description = "";
type = with types; listOf anything;
default = [];
};
files = mkOption {
type = with types; listOf anything;
default = [];
};
persistDirectory = mkOption {
default = "/persist";
type = types.str;
description = ''
The directory in which persistent files live.
'';
};
persistGroupName = mkOption {
default = "persist";
type = types.str;
};
files = mkOption {
description = "";
type = with types; listOf anything;
default = [];
};
device = mkOption {
description = "Device to be wiped";
type = types.str;
};
archiveLimit = mkOption {
type = types.ints.positive;
description = "The number of previous roots to preserve.";
default = 3;
description = ''
Name of the group whose members have access to the persist directory.
'';
};
};
};
@@ -51,115 +47,19 @@ in {
name = cfg.persistGroupName;
};
systemd.tmpfiles.settings = {
"10-persist" = {
# Permit members of `cfg.persistGroupName` to read, write, and execute
# /persist.
"/persist" = {
z = {
group = cfg.persistGroupName;
mode = "2775";
};
};
# Permit members of `cfg.persistGroupName` to read, write, and execute
# /persist.
systemd.tmpfiles.settings."10-persist".${cfg.persistDirectory} = {
z = {
group = cfg.persistGroupName;
mode = "2775";
};
};
# TODO: Move this somewhere else.
programs.fuse.userAllowOther = true;
# sudo mount -o ro -t virtiofs mount-dots /persist/dots
boot.initrd.systemd = {
# `pkgs.babashka` calls `pkgs.babashka-unwrapped`, and will explode if it
# cannot find it.
storePaths = [ pkgs.babashka-unwrapped ];
initrdBin = with pkgs; [
babashka
util-linux
];
};
boot.initrd.systemd.services.erase-darlings =
let service = {
description = "Rollback filesystem to a blank state on boot";
wantedBy = [
"initrd.target"
];
before = [
"sysroot.mount"
];
path = with pkgs; [ util-linux babashka ];
unitConfig.DefaultDependencies = "no";
serviceConfig = {
Type = "oneshot";
RemainAfterExit = true;
};
script =
let bb-script = "/sysroot" + ./impermanence/erase-darlings.clj;
in ''${pkgs.babashka}/bin/bb "${bb-script}" -n "${toString cfg.archiveLimit}" --device "${cfg.device}"'';
};
in if config.boot.initrd.systemd.enable
then service
else throw "sydnix.impermanence currently requires config.boot.initrd.systemd.enable'!";
# TODO: Remove this.
# boot.initrd.systemd.services.erase-darlings-old =
# let service = {
# description = "Rollback filesystem to a blank state on boot";
# wantedBy = [
# "initrd.target"
# ];
# after = [
# # "zfs-import.service"
# "zfs-import-rpool.service"
# ];
# before = [
# "sysroot.mount"
# ];
# path = [ pkgs.zfs ];
# unitConfig.DefaultDependencies = "no";
# serviceConfig = {
# Type = "oneshot";
# RemainAfterExit = true;
# };
# script = /* bash */ ''
# zfs rollback -r rpool/local/root@blank \
# && echo ">> >> rollback complete << <<"
# '';
# };
# in if config.boot.initrd.systemd.enable
# then service
# else throw "sydnix.impermanence currently requires config.boot.initrd.systemd.enable'!";
# systemd.services =
# let erase-home-darlings = {
# description = "Rollback home to a blank state on boot";
# wantedBy = [
# "local-fs-pre.target"
# "zfs-mount.service"
# ];
# before = [
# "local-fs.target"
# "local-fs-pre.target"
# "zfs-mount.service"
# ];
# path = [ pkgs.zfs pkgs.babashka pkgs.util-linux ];
# unitConfig.DefaultDependencies = "no";
# serviceConfig = {
# Type = "oneshot";
# RemainAfterExit = true;
# ExecStart =
# let script = ./erase-home-darlings.clj;
# in ''${pkgs.babashka}/bin/bb "${script}" -n "${toString cfg.archiveLimit}" --dataset "${cfg.dataset}" --rollback-to "${cfg.rollbackTo}"'';
# };
# stopIfChanged = false;
# restartIfChanged = false;
# };
# in {
# # inherit erase-home-darlings;
# };
environment.persistence."/persist/root" = {
environment.persistence."${cfg.persistDirectory}/root" = {
directories = cfg.directories;
files = cfg.files;
};

View File

@@ -1,43 +0,0 @@
#!/usr/bin/env bb
(require '[clojure.core.match :refer [match]]
'[babashka.cli :as cli]
'[clojure.pprint :as pp]
'[clojure.tools.logging :as l]
'[babashka.process :refer [shell check process] :as p])
(def cli-spec
{:spec
{:archive-limit {:coerce :int
:alias :n
:validate #(pos? %)
:default 3
:desc "Number of archives to save at a time."}
:device {:coerce :string
;; :validate fs/exists?
:require true
:desc "Dataset to be archived and rolled back."}
:previous-roots
{:coerce :string
:default "/persist/previous"
:desc "The path under which previous roots will be stored."}
:error-fn
(fn [{:keys [spec type cause msg option] :as data}]
(when (= :org.babashka/cli type)
(case cause
:require
(println
(format "Missing required argument: %s\n" option))))
(System/exit 1))}})
(defmacro with-echoed-shell-commands [& body]
(let [print-cmd #(println (str "+ " (str/join (:cmd %))))]
`(binding [p/*defaults* {:pre-start-fn ~print-cmd}]
~@body)))
(defn -main [opts]
(pp/pprint opts)
(with-echoed-shell-commands
(shell "mount")))
(-main (cli/parse-opts *command-line-args* cli-spec))

View File

@@ -0,0 +1,59 @@
{ config, lib, pkgs, ... }:
with lib;
let cfg = config.impermanence.rollback;
in {
options = {
impermanence.rollback = {
enable = mkEnableOption "rollback of the root filesystem";
device = mkOption {
description = "Device to be wiped";
type = types.str;
};
subvolume = mkOption {
description = ''
The subvolume to be rolled back. Usually, this should be where your
root filesystem resides.
'';
type = types.str;
};
};
};
config = mkIf cfg.enable {
boot.initrd.postDeviceCommands = ''
# Mount the btrfs filesystem.
mkdir -p /btrfs-tmp
mount -t btrfs "${cfg.device}" /btrfs-tmp
# If the moribound subvolume exists, send it do 'death row' (old-roots),
# where live for about three days before its eventual deletion.
if [[ -e "/btrfs-tmp/${cfg.subvolume}" ]]; then
mkdir -p /btrfs-tmp/old-roots
timestamp=$(date --date="@$(stat -c %Y "/btrfs-tmp/${cfg.subvolume}")" "+%Y-%m-%-d_%H:%M:%S")
mv "/btrfs-tmp/${cfg.subvolume}" "/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"
}
# Delete previous roots older than three days.
# TODO: I would prefer archiving the last N previous roots, rather than
# time.
for i in $(find /btrfs-tmp/old-roots/ -maxdepth 1 -mtime +3); do
delete_subvolume_recursively "$i"
done
btrfs subvolume create "/btrfs-tmp/${cfg.subvolume}"
umount /btrfs-tmp
'';
};
}