Lab: Disk identification and partition layout

This lab walks through the storage on a real host, from the mount point down to the physical disk, and designs a partition scheme for a new disk.

Objective

By the end of this lab, you can:

Map every mount point to its physical disk.
Verify that /etc/fstab uses stable identifiers.
Design a partition scheme for a new disk and justify each choice.
Demonstrate the safe use of wipefs to repurpose a disk.

Architecture

The host has these storage components:

A boot disk (sda) with a GPT partition table.
A data disk (sdb) with one large ext4 partition.
An optional NVMe device (nvme0n1) used for fast data.

Tasks

Task 1: Map every mount point to its physical disk

For each mount point, trace from mount through filesystem through LV/partition to disk:

df -h
lsblk -o name,size,fstype,mountpoint
findmnt /
lvs -o+devices vg0

Write down the chain for each mount: e.g. /var -> vg0-var -> /dev/sda2 partition 2 -> /dev/sda (SATA port 1).

Task 2: Verify /etc/fstab uses stable identifiers

cat /etc/fstab
for fstab_entry in $(grep -v '^#\|^$' /etc/fstab | awk '{print $1}'); do
    if [[ "$fstab_entry" =~ ^UUID= ]]; then
        echo "OK: $fstab_entry uses UUID"
    elif [[ "$fstab_entry" =~ ^LABEL= ]]; then
        echo "OK: $fstab_entry uses LABEL"
    else
        echo "WARNING: $fstab_entry does NOT use a stable identifier"
    fi
done

If any entry uses /dev/sd* or other device names, the host will not boot after a hardware change. Fix the entry and run mount -a to verify.

Task 3: Identify a new disk and design a partition scheme

If a new disk is available (loop device or attached USB):

lsblk -d -o name,size,tran,model

Design a partition scheme with rationale:

Partition 1: small, FAT32, EFI System Partition (if the host uses UEFI).
Partition 2: rest of disk, ext4 or xfs, for data.
OR partition 2: LVM PV for flexible storage.

Write down the fstab entry you would use, with the UUID you expect to find after mkfs.

Task 4: Demonstrate safe wipefs usage

If a candidate disk is available (a loop device created from a spare file is fine):

# Create a small file to use as a loop device
truncate -s 1G /tmp/safety-disk.img
LOOP=$(sudo losetup --find --show /tmp/safety-disk.img)
echo "Loop device: $LOOP"

# Give it a signature to find. A blank file has none.
sudo mkfs.ext4 -q "$LOOP"

# Inspect signatures before wiping
sudo blkid "$LOOP"        # TYPE="ext4" UUID="..."

# List mode: no options, nothing is erased
sudo wipefs "$LOOP"       # prints offset / uuid / type / label

# Real dry run of the erase: -n with -a
sudo wipefs -n -a "$LOOP"

# Now actually wipe. -a erases ALL signatures; -b writes a
# restorable backup per signature into $HOME.
sudo wipefs -a -b "$LOOP"
sudo blkid "$LOOP"        # now returns nothing, exit status 2
ls ~/wipefs-*.bak         # the undo file

# Clean up
sudo losetup -d "$LOOP"
rm /tmp/safety-disk.img
rm -f ~/wipefs-*.bak

Two flags carry this task.

-a is what erases. Bare wipefs DEVICE is list mode — the man page says so plainly — and it prints a tidy table of signatures that reads exactly like a confirmation of work done. It is not. Without -a, blkid afterwards still reports the original TYPE=, and the step above would fail its own check.

-b is what makes the erase reversible. It writes ~/wipefs-<devname>-<offset>.bak for each signature it removes. Restore with:

sudo dd if=~/wipefs-sdb-0x00000438.bak of=/dev/sdb \
  seek=$((0x00000438)) bs=1 conv=notrunc

Without -b there is no undo. Type both flags every time.

The dry run is wipefs -n -a, not wipefs --no-act on its own: --no-act suppresses the write, but with no -a there was never going to be a write to suppress.

Validation

Every mount point in the host maps to a physical disk via a traceable chain.
/etc/fstab has zero entries using /dev/sd* device names (unless the disk is a removable device that is always inserted the same way, which is rare).
The new-disk design has a written rationale for each partition.
The wipefs walkthrough produced a clear before/after blkid output: TYPE="ext4" before, no output after. If blkid still reports a type, -a was omitted and nothing was erased.
ls ~/wipefs-*.bak lists at least one backup file, proving the erase was reversible.

Cleanup

If you used a loop device for Task 4, the losetup -d and rm steps above clean up. The fstab changes from Task 2 may need to be reverted if you did not commit them with intent (use git diff /etc/fstab to inspect).

What you learned

The host’s storage can be traced from mount point to physical disk through a chain of layers (filesystem, LV, PV, partition, device).
/etc/fstab should always use UUID or LABEL. /dev/sd* names are not stable.
Partition design starts with the disk’s role (boot, data, swap) and ends with a partition scheme.
wipefs has two modes. No options is a read-only listing; -a is the one that erases. Reading the listing as proof of a wipe is a common and expensive mistake.
wipefs -a -b writes a per-signature backup to $HOME, so the erase can be undone with dd. Without -b there is no undo.
The genuine dry run is wipefs -n -a, and it is the safety net.

Objectives

Prerequisites