Planning a disk layout

What is worth its own partition, and why

The reason to separate a directory is almost always that something can fill it. /var holds logs, mail spools and package caches, and a runaway log on a single-partition system fills the root filesystem and stops the machine. /home holds user data, and separating it means a reinstall can keep it. /tmp is world-writable, so a separate /tmp bounds what any user can consume and lets it be mounted with noexec and nosuid.

/boot is separated for a different reason: something has to be readable before the machine is fully up. If the root filesystem is encrypted or sits on an exotic volume manager, the kernel and initramfs have to live somewhere the boot loader can read unaided. On UEFI systems there is also an EFI system partition, formatted FAT and usually mounted at /boot/efi, which the firmware itself reads.

Swap is not a directory but a partition or file the kernel uses to page memory out. The old rule of twice RAM is long obsolete; what matters for the exam is that swap exists, that it is type 82 on an MBR table, that mkswap prepares it and swapon activates it, and that hibernation needs swap at least the size of RAM.

LVM, and mount points as a layer of indirection

A classic partition table pins sizes at install time and resizing them later means moving data. LVM adds a layer: physical volumes (the partitions or disks) are grouped into a volume group, and logical volumes are carved out of that group and can be grown, shrunk or moved while the system runs. The filesystem sits on the logical volume, so /home can borrow space from the volume group next year without repartitioning.

The three layers have three sets of commands named after them, and the exam asks for them by name rather than by description. pvcreate initialises a partition or a whole disk as a physical volume; vgcreate pools one or more physical volumes into a named volume group; lvcreate carves a logical volume out of that group, and only the logical volume ever receives a filesystem and a mount point. Growing follows the same naming one word further on — vgextend adds another physical volume to an existing group, lvextend hands group space to a logical volume — and the filesystem on top is a separate step again, grown with resize2fs or xfs_growfs after the volume beneath it has moved. pvs, vgs and lvs report each layer.

Either way, the thing a user sees is a mount point — a directory in the single tree at which a filesystem is attached. Linux has no drive letters: a second disk is not "D:", it is whatever directory you mount it on, and the same filesystem mounted somewhere else is the same data at a different path.