Creating a btrfs RAID 1 array for Debian 12/13

April 7, 2025

The debian installer doesn’t support creating a RAID 1 btrfs array directly, so you must do it manually. Follow the debian installer instructions to install debian with a ESP partition and BTRFS partition. For UEFI systems you don’t need the EFI partitions to be raided but for BIOS systems you will need to create a mdadm RAID array for the boot partition.

I am installing the BTRFS array to nvme1 and nvme2 because my windows install is on nvme0. I installed debian to nvme1 with a single 1GB ESP partition and a 1.9T BTRFS partition with nvme2 as the second disk for the RAID 1 array.

edward@edward:~$ lsblk
NAME        MAJ:MIN RM   SIZE RO TYPE MOUNTPOINTS
nvme1n1     259:0    0   1.9T  0 disk 
├─nvme1n1p1 259:3    0   953M  0 part /boot/efi
└─nvme1n1p2 259:4    0   1.9T  0 part /
nvme2n1     259:1    0   1.9T  0 disk

I then needed to replicate the partitions from the installed disk to the new disk.

Note

The -G flag is important as it randomises the partuuid’s of the cloned partitions.

edward@edward:~$ sgdisk /dev/nvme1n1 -R /dev/nvme2n1
edward@edward:~$ sgdisk -G /dev/nvme2n1
edward@edward:~$ lsblk -f
NAME        FSTYPE FSVER LABEL                   UUID                                 FSAVAIL FSUSE% MOUNTPOINTS
nvme1n1                                                                                              
├─nvme1n1p1 vfat   FAT32                         AD20-7BC4                             945.3M     1% /boot/efi
└─nvme1n1p2 btrfs                                cad8f221-dbb3-4e78-b0d9-f5f04ec43c4f    1.6T    11% /
nvme2n1                                                                                              
├─nvme2n1p1                                                                                 
└─nvme2n1p2

I then created a vfat filesystem on the new ESP partition of nvme2.

edward@edward:~$ mkfs.vfat /dev/nvme2n1p1
mkfs.fat 4.2 (2021-01-31)

edward@edward:~$ lsblk -f
NAME        FSTYPE FSVER LABEL                   UUID                                 FSAVAIL FSUSE% MOUNTPOINTS
nvme1n1                                                                                              
├─nvme1n1p1 vfat   FAT32                         AD20-7BC4                             945.3M     1% /boot/efi
└─nvme1n1p2 btrfs                                cad8f221-dbb3-4e78-b0d9-f5f04ec43c4f    1.6T    11% /
nvme2n1                                                                                              
├─nvme2n1p1 vfat   FAT32                         A172-F26A                                           
└─nvme2n1p2

Then i needed to create a new mount for the secondary ESP partition and modified the fstab file to mount the new ESP partition at /boot/efi2. I also added the degraded flag to the btrfs mount to allow booting even with a disk failure. The same flag for the ESP partitions are for the same reason.

edward@edward:~$ mkdir /boot/efi2

/etc/fstab

# / was on /dev/nvme1n1p2 during installation
UUID=cad8f221-dbb3-4e78-b0d9-f5f04ec43c4f /		btrfs   defaults,subvol=@rootfs,degraded 0      0
# /boot/efi was on /dev/nvme1n1p1 during installation
UUID=AD20-7BC4  /boot/efi       vfat    umask=0077,nofail      0    1
UUID=A172-F26A	/boot/efi2	vfat	umask=0077,nofail	0   1

I then installed a grub script to sync the two boot partitions whenever the EFI is modified or updated.

/etc/grub.d/90_sync_boot

#!/bin/sh
if mountpoint -q "/boot/efi" && mountpoint -q "/boot/efi2"; then
	rsync -av --stats --delete-delay --delay-updates /boot/efi/ /boot/efi2/
fi

edward@edward:~$ sudo chmod +x /etc/grub.d/90_sync_boot
edward@edward:~$ sudo update-grub

Then i added the 2nd partiton of nvme2 to the BTRFS array.

edward@edward:~$ sudo btrfs device usage /
/dev/nvme1n1p2, ID: 1
   Device size:             1.86TiB
   Device slack:              0.00B
   Data,single:           231.01GiB
   Metadata,DUP:            6.00GiB
   System,DUP:             64.00MiB
   Unallocated:             1.63TiB

edward@edward:~$ sudo btrfs device add /dev/nvme2n1p2 /
Performing full device TRIM /dev/nvme2n1p2 (1.86TiB) ...

edward@edward:~$ sudo btrfs device usage /
/dev/nvme1n1p2, ID: 1
   Device size:             1.86TiB
   Device slack:              0.00B
   Data,single:           231.01GiB
   Metadata,DUP:            6.00GiB
   System,DUP:             64.00MiB
   Unallocated:             1.63TiB

/dev/nvme2n1p2, ID: 2
   Device size:             1.86TiB
   Device slack:              0.00B
   Unallocated:             1.86TiB

Then i had to convert the filesystem data and metadata from single to RAID 1.

edward@edward:~$ sudo btrfs balance start --full-balance -dconvert=raid1 -mconvert=raid1 /
Done, had to relocate 236 out of 236 chunks

edward@edward:~$ sudo btrfs device usage /
/dev/nvme1n1p2, ID: 1
   Device size:             1.86TiB
   Device slack:              0.00B
   Data,RAID1:            233.00GiB
   Metadata,RAID1:          4.00GiB
   System,RAID1:           32.00MiB
   Unallocated:             1.63TiB

/dev/nvme2n1p2, ID: 2
   Device size:             1.86TiB
   Device slack:              0.00B
   Data,RAID1:            233.00GiB
   Metadata,RAID1:          4.00GiB
   System,RAID1:           32.00MiB
   Unallocated:             1.63TiB

Then i scrubbed the filesystem just in case.

edward@edward:~$ sudo btrfs scrub start -Bd /
Scrub device /dev/nvme1n1p2 (id 1) done
Scrub started:    Fri Dec  6 20:24:47 2024
Status:           finished
Duration:         0:01:08
Total to scrub:   237.03GiB
Rate:             3.14GiB/s
Error summary:    no errors found

Scrub device /dev/nvme2n1p2 (id 2) done
Scrub started:    Fri Dec  6 20:24:47 2024
Status:           finished
Duration:         0:01:08
Total to scrub:   237.03GiB
Rate:             3.14GiB/s
Error summary:    no errors found

I then tested that the RAID array works with only a single disk by removing a disk and booting from the remaining disk. Now my debian system now has redundant storage.