Understanding storage¶
Before a disk can store files, it goes through four layers: the disk itself, a partition table that divides it, a file system on each partition, and a mount point where the file system appears in the directory tree.
Disks and partitions¶
Linux shows every disk as a file in /dev:
| Name | Typical device |
|---|---|
/dev/sda, /dev/sdb |
SATA and SCSI disks, also most virtual disks |
/dev/vda |
Virtual disks with the virtio driver (KVM, many cloud providers) |
/dev/nvme0n1 |
NVMe SSDs |
/dev/sda1, /dev/nvme0n1p1 |
The first partition on that disk |
/dev/mapper/vg_data-lv_srv |
A logical volume of LVM |
On the example server tuxdex-srv, the system disk sda has three partitions, the
backup disk sdb one, and the disk sdc belongs to LVM.
Partition tables¶
| Type | Maximum disk size | Partitions | Use |
|---|---|---|---|
| GPT | practically unlimited | 128 | Standard today, required for disks larger than 2 TiB and for UEFI |
| MBR (DOS) | 2 TiB | 4 primary | Old systems and old BIOS boot |
Use GPT for new disks.
File systems¶
| File system | Strengths | Grow | Shrink |
|---|---|---|---|
| ext4 | Robust, universal, the classic on Linux | yes, also while mounted | yes, only unmounted |
| XFS | Very good with large files and many parallel accesses | yes, while mounted | no |
| Btrfs | Snapshots, checksums, compression, several disks | yes | yes |
| vfat | Readable everywhere, required for the EFI partition | rarely | rarely |
| swap | Not a file system: swap space for memory |
/etc/fstab: mounting at boot¶
Every line in /etc/fstab describes one file system that is mounted at boot:
# <file system> <mount point> <type> <options> <dump> <pass>
UUID=c1f4e2a7-5b3d-4e8a-9f61-2d7b8c0e9a14 / ext4 errors=remount-ro 0 1
UUID=4A1C-7F32 /boot/efi vfat umask=0077 0 1
UUID=8e2b6d4f-1a3c-4f5e-b7d9-0c2e4a6f8b31 none swap sw 0 0
UUID=87c6aa0c-cb18-455e-b534-d2864a90bb80 /mnt/backup ext4 defaults,nofail 0 2
/dev/mapper/vg_data-lv_srv /srv/data ext4 defaults 0 2
| Field | Meaning |
|---|---|
| File system | Which device. Use UUID= from blkid, because names like sdb can change |
| Mount point | Where it appears in the directory tree |
| Options | defaults, plus e.g. nofail (boot even if the disk is missing) or noatime |
| Pass | Order of the check at boot: 1 for /, 2 for the rest, 0 never |
LVM: flexible volumes¶
LVM puts a layer between disks and file systems:
| Layer | Name | Command |
|---|---|---|
| Disks or partitions | Physical volumes (PV) | pvs, pvcreate |
| Pool from one or more PVs | Volume groups (VG) | vgs, vgcreate, vgextend |
| "Partitions" taken from the pool | Logical volumes (LV) | lvs, lvcreate, lvextend |
The big advantage: when a volume gets full, you add a disk to the volume group and grow the logical volume, while the server keeps running.
RAID: redundancy¶
A software RAID (mdadm) combines several disks into one device /dev/md0:
| Level | Disks | Usable | Survives |
|---|---|---|---|
| RAID 1 (mirror) | 2 or more | one disk | all but one disk failing |
| RAID 5 | 3 or more | all but one | one failed disk |
| RAID 6 | 4 or more | all but two | two failed disks |
| RAID 10 | 4 or more | half | one disk per mirror pair |
On top of /dev/md0 you create a file system or an LVM physical volume, just as on a
normal disk. RAID protects against failed hardware, not against deleted files.
Distribution differences¶
| Debian | Ubuntu | RHEL | Arch | |
|---|---|---|---|---|
| Default file system of the installer | ext4 | ext4 | XFS | chosen during installation |
| LVM in the default installation | optional | yes | yes | no |
/tmp in memory (tmpfs) |
yes, since Debian 13 | no | no | yes |
UUIDs instead of device names
The order of disks can change after a reboot or when a disk is added: yesterday's
sdb can be sdc today. UUIDs in /etc/fstab always point to the right file system.
Test /etc/fstab before rebooting
A wrong line in /etc/fstab can stop the boot process. After every change, run
sudo findmnt --verify and sudo mount -a. Only reboot if both report no errors.