Skip to content

Commit

Permalink
Update table-to-gpt migration guide
Browse files Browse the repository at this point in the history
- Generalize examples to be independent of ZFS
- Reduce indent of examples to focus on the relevant bits
- Add explicit migration steps, pre- and postconditions
  • Loading branch information
mweinelt authored and mergify[bot] committed May 31, 2024
1 parent 9d5c673 commit 0274af4
Showing 1 changed file with 98 additions and 88 deletions.
186 changes: 98 additions & 88 deletions docs/table-to-gpt.md
Original file line number Diff line number Diff line change
@@ -1,117 +1,127 @@
# disko table to gpt migration guide
NixOS on ZFS with latest disko
## Error
# Migrating to the new GPL layout

WHen deploying NixOS the following trace occurs
## Situation

```clean=
When evaluating your NixOS system closure the following trace appears:

```
trace: warning: The legacy table is outdated and should not be used. We recommend using the gpt type instead.
Please note that certain features, such as the test framework, may not function properly with the legacy table type.
If you encounter errors similar to:
"error: The option `disko.devices.disk.disk1.content.partitions."[definition 1-entry 1]".content._config` is read-only, but it's set multiple times,"
this is likely due to the use of the legacy table type.
```

## Before
Configuration in question
The solution is to migrate to the new `gpt` layout type.

## Precondition

Disko was set up with

```nix=
{ disk ? "/dev/nvme0n1", ... }:
- `type = "table"` and
- `format = "gpt"`,

for example like this:

```nix
{
boot.supportedFilesystems = [ "zfs" ];
disko.devices = {
disk = {
nvme = {
type = "disk";
device = disk;
content = {
type = "table";
format = "gpt";
partitions = [
{
name = "ESP";
start = "0";
end = "512MiB";
fs-type = "fat32";
bootable = true;
content = {
type = "filesystem";
format = "vfat";
mountpoint = "/boot";
};
};
{
name = "zfs";
start = "512MiB";
end = "100%";
content = {
type = "zfs";
pool = "tank";
};
}
];
};
};
};
zpool = {
...
disko.devices.disk.example = {
type = "disk";
device = "/dev/nvme0n1";
content = {
type = "table";
format = "gpt";
partitions = [
{
name = "ESP";
start = "0";
end = "512MiB";
fs-type = "fat32";
bootable = true;
content = {
type = "filesystem";
format = "vfat";
mountpoint = "/boot";
};
}
{
name = "root";
start = "512MiB";
end = "100%";
content.format = "ext4";
}
];
};
};
}
```
# Migration

## Remediation

The new GPT layout (`type = "gpt"`) uses partlabels to realize the partiton numbering. For this reason you have to manually set up partition labels, if you want to resolve this issue.

### Create GPT partition labels

For each partition involved, create the partition label from these components:

- The partition number (e.g. /dev/nvme0n**1**, or /dev/sda**1**)
- The parent type in your disko config (value of `disko.device.disk.example.type = "disk";`)
- The parent name in your disko config (attribute name of `disko.devices.disk.example`, so `example` in this example)
- The partition name in your disko config (attribute name of `disko.devices.disk.content.partitions.*.name`)

The new gpt layout uses partlabels to unify the partiton numbering. for this reason you have to set the partition labels manually if you want to upgrade.
```bash
sgdisk -c 1:disk-nvme-ESP /dev/nvme0n1
sgdisk -c 2:disk-nvme-zfs /dev/nvme0n1
# sgdisk -c 1:disk-example-ESP /dev/nvme0n1
# sgdisk -c 2:disk-example-zfs /dev/nvme0n1
Warning: The kernel is still using the old partition table.
The new table will be used at the next reboot or after you
run partprobe(8) or kpartx(8)
The operation has completed successfully.
```

1 is the partition number (in system `/dev/nvme0n1p`**`1`**)
disk is the parents type (in config: `disko.devices.disk.nvme.type = `**`"disk"`**)
nvme is the parents name ( in config: `disko.devices.disk.`**`nvme`**)
ESP is the name of the partition (in config: `disko.devices.disk.content.partitions.`**`ESP`**)

usually it's okay to fuck this up, since booting an old generation should still be possible as the previous setup used disk numbering instead. When booting the new system you will see that systemd is waiting for the device with the respective partlabel.
# After

```nix=
disko.devices = {
disk = {
nvme = {
type = "disk";
device = disk;
content = {
type = "gpt";
partitions = {
ESP = {
size = "512MiB";
type = "EF00";
priority = 1; # instead of "start" the priority is used, otherwise the partitions are created alphabetically (ESP before zfs)
content = {
type = "filesystem";
format = "vfat";
mountpoint = "/boot";
};
};
zfs = {
size = "100%";
content = {
type = "zfs";
pool = "tank";
};
};
### Update disko configuration

Make the following changes to your disko configuration:

1. Set `disko.devices.disk.example.content.type = "gpt"`
1. Remove `disko.devices.disk.example.format`
1. Convert `disko.devices.disk.example.partitions` to an attribute set and promote the `name` field to the key for its partition
1. Add a `priority` field to each partition, to reflect the intended partition number

Then rebuild your system and reboot.

### Recovering from mistake

If you made a mistake here, your system will be waiting for devices to appear, and then run into timeouts. You can easily recover from this, since rebooting into an old generation will still use the legacy way of numbering of partitions.

## Result

The fixed disko configuration would look like this:

```nix
{
disko.devices.disk.example = {
type = "disk";
device = "/dev/nvme0n1";
content = {
type = "gpt";
partitions = {
ESP = {
size = "512MiB";
type = "EF00";
priority = 1;
content = {
type = "filesystem";
format = "vfat";
mountpoint = "/boot";
};
};
root = {
size = "100%";
priority = 2;
content.format = "ext4";
};
};
};
zpool = { ... };
}
};
}
```

0 comments on commit 0274af4

Please sign in to comment.