Partitions and filesystems

Two partition schemes

The MBR scheme keeps its table in the first sector and allows four primary partitions; getting more means turning one of them into an extended partition that contains logical ones. It cannot address disks beyond 2 TB. GPT replaces it with a much larger table, keeps a backup copy at the end of the disk, and has no primary/extended distinction — which is why every modern install uses it, and why questions about the four-partition limit are always about MBR.

fdisk edits MBR tables (and, in current versions, GPT too), gdisk is the GPT-specific equivalent, and parted handles both and can also resize. A partition carries a type id: 83 is a Linux filesystem, 82 is Linux swap, and on GPT there is an EFI system partition type that the firmware looks for. Changing a type id does not change what is on the partition — it is a label, and a wrong one merely misleads.

fdisk -l is the non-interactive half of the same tool and the quickest inventory of what a machine has: given -l, or the long form --list, it prints the partition table of every disk it can find and exits without ever offering its editing prompt. Naming a disk restricts the report to it. Because it reads the tables by opening the block devices directly, an ordinary user gets nothing back at all, which is why it is normally run as sudo fdisk -l and why "it printed nothing" is a permissions answer rather than an empty machine.

Filesystems, and picking one

mkfs creates a filesystem, dispatching to a type-specific tool: mkfs.ext4, mkfs.xfs, mkfs.vfat, or mkfs -t ext4 spelled the other way. ext4 is the general-purpose default across most distributions; ext3 is its predecessor and ext2 the version before journaling, which is why ext2 is still occasionally chosen for small /boot partitions and USB sticks. XFS is the default on Red Hat family systems and is strong on large files; it can be grown but not shrunk. Btrfs adds snapshots and its own volume management, and that second half changes how it is created: mkfs.btrfs accepts a LIST of devices and stripes or mirrors across them itself, with -d choosing the profile for data and -m the profile for metadata, so mkfs.btrfs -d raid1 -m raid1 /dev/sdc /dev/sdd builds a mirrored filesystem with no md array or LVM layer beneath it. vfat exists for interoperability, which is why the EFI system partition uses it, and mkfs.fat (formerly mkfs.vfat) chooses FAT12, FAT16 or FAT32 from the partition size unless you say otherwise.

Behind all three ext filesystems there is one program, and it is worth knowing by name because a question will use it instead of the friendly spelling. mke2fs is the real tool; mkfs.ext2, mkfs.ext3 and mkfs.ext4 are the same binary deciding what to build from the name it was called by, and mke2fs -t ext4 /dev/sdb1 is exactly mkfs.ext4 /dev/sdb1. mkfs itself is only a front end that appends the type to its own name and runs whatever it finds, which is why a filesystem with no mkfs.<type> installed fails with "command not found" rather than an error about the filesystem.

Swap is prepared with mkswap and enabled with swapon; swapon -s or /proc/swaps shows what is active, and an entry in /etc/fstab makes it come back after a reboot. A swap FILE works as well as a partition and is created with dd or fallocate, then given to mkswap the same way.