Disks and Mounts: lsblk, fdisk, mkfs and fstab

Adding storage is routine: a database needs more space, logs need their own volume, or a cloud provider hands you a new unattached disk. The steps are always the same: identify, partition, format, mount, make it permanent. The danger is working on the wrong device, so every step below checks before it changes anything.

Versions in use on this date. Ubuntu 18.04.3 LTS (released 8 August 2019) with util-linux 2.31 and e2fsprogs 1.44. CentOS 7.6 uses XFS by default for new installs; the commands below are for ext4, with the XFS equivalents noted.

From raw disk to mounted storage

flowchart TB
  D[New disk /dev/sdb<br/>raw block device] --> I[Identify with lsblk]
  I --> PT[Partition table<br/>fdisk or parted]
  PT --> PA[Partition /dev/sdb1]
  PA --> FS[mkfs.ext4 creates filesystem]
  FS --> MT[mount to /data]
  MT --> UU[Record UUID with blkid]
  UU --> FT[Add line to /etc/fstab]
  FT --> TS[Test with mount -a]
  TS --> OK[Survives reboot]

1. Identify the right disk

lsblk -f                                # tree with filesystems, UUIDs, mount points
sudo fdisk -l /dev/sdb                  # size and any existing partition table
sudo blkid /dev/sdb

The new disk has no partitions and no FSTYPE. If lsblk shows data or a mount point, stop: it is not the empty disk you think it is. In cloud servers, confirm by size, since device names (/dev/sdb, /dev/xvdf, /dev/nvme1n1) can differ between instances.

2. Partition

sudo fdisk /dev/sdb
#   n (new)  p (primary)  1  Enter  Enter   w (write)
sudo partprobe /dev/sdb
lsblk /dev/sdb

fdisk changes nothing until you press w; q quits safely. Disks over 2 TB need GPT (sudo parted /dev/sdb mklabel gpt). For a data disk you may also skip partitioning and format the whole device, but a partition table is clearer to the next engineer.

3. Create the filesystem

sudo mkfs.ext4 -L data /dev/sdb1                # ext4 with label "data"
# CentOS default alternative:
# sudo mkfs.xfs -L data /dev/sdb1
sudo tune2fs -m 1 /dev/sdb1                     # reduce reserved blocks from 5% to 1% on a data disk

mkfs destroys existing data on that device with no confirmation prompt. Double-check the device name.

4. Mount, then make it permanent

sudo mkdir -p /data
sudo mount /dev/sdb1 /data
df -hT /data
sudo blkid /dev/sdb1                            # note the UUID

Edit /etc/fstab using the UUID, not /dev/sdb1, because device names can change between boots:

UUID=3f2c1d0e-xxxx-xxxx-xxxx-xxxxxxxxxxxx  /data  ext4  defaults,noatime,nofail  0  2

Fields: device, mount point, type, options, dump (0), fsck order (1 for /, 2 for others, 0 for none). nofail lets the system boot even if the disk is missing, which matters for data disks on cloud servers.

Always test before rebooting:

sudo umount /data
sudo mount -a              # mounts everything in fstab; errors appear now, not at boot
findmnt /data

A bad fstab line can drop the machine into emergency mode at the next reboot. Keep a copy first: sudo cp /etc/fstab /etc/fstab.bak.

Permissions and ownership

A new ext4 filesystem is owned by root. Give the application what it needs:

sudo chown -R postgres:postgres /data/pgsql
sudo chmod 750 /data/pgsql

Checking health and capacity

df -hT                      # space; df -i shows inodes
sudo du -xh --max-depth=1 /data | sort -h | tail
sudo fsck -n /dev/sdb1      # read-only check (unmounted only)
sudo smartctl -H /dev/sda   # (apt install smartmontools) physical disk health
iostat -x 5 3               # (apt install sysstat) utilisation and latency

Alert before a filesystem reaches 80 to 85% full. Growing a disk without downtime is covered later in this series with LVM.

Pitfalls

• Mount hides existing files: mounting onto a non-empty directory covers its content until you unmount.
• fstab typo causes boot failure: use mount -a first and nofail for non-critical mounts.
• Wrong device formatted: always lsblk immediately before mkfs.
• Full inodes: millions of tiny files exhaust inodes while space remains; df -i shows it.