fstab syntax - fields, options, and the dump/pass columns · Linux

/etc/fstab lists the filesystems the host should mount. An error in it either prevents the host from booting or causes the affected filesystem to fail to mount.

Nothing in the kernel reads this file. On a systemd host, systemd-fstab-generator parses /etc/fstab early in boot and generates a .mount unit for each entry; systemd then orders and mounts them through the normal unit dependency graph, and systemd-fsck@.service runs the check. mount -a reads the file directly, but that is the mount command, not the kernel either.

That matters for troubleshooting, because it tells you where to look:

systemctl list-units --type=mount           # what was generated, and its state
systemctl cat data.mount                    # the unit generated from your entry
journalctl -u systemd-fstab-generator -b    # parse errors in the file itself

It also explains why the pass column is not what orders your mounts. Ordering comes from the unit dependencies systemd derives from the mount points (a mount at /var/log is ordered after /var because the path nests), while the sixth column only sets the fsck pass number.

The six columns

Read-only / Safefstab columns— Six whitespace-separated columns: filesystem, mount point, type, options, dump, pass. UUIDs are used to identify the device; mount point is where the filesystem attaches; type is the filesystem; options are comma-separated; dump is a legacy dump field (set to 0); pass is the fsck order (0=skip, 1=root first, 2=after root).

$ cat /etc/fstab

UUID=1a2b-3c4d... /boot ext4 defaults 0 2
UUID=5e6f-7a8b... / ext4 defaults,noatime 0 1
UUID=9a8b-7c6d... /var ext4 defaults 0 2
UUID=f1e2-d3c4... swap swap defaults 0 0

Column	Meaning
1. Filesystem	UUID, LABEL, PARTUUID, or device path
2. Mount point	Absolute path; `none` for swap
3. Filesystem type	`ext4`, `xfs`, `btrfs`, `swap`, `nfs`, etc.; `auto` to let the kernel detect
4. Mount options	Comma-separated: `defaults`, `noatime`, `errors=remount-ro`, etc.
5. Dump field	Legacy `dump(8)` flag: 0 = skip, 1 = include. Use 0.
6. Pass field	`fsck` order: 0 = skip, 1 = root first, 2 = after root.

Common mount options

Read-only / Safemount output— mount shows the current effective options. Per man 8 mount, `defaults` expands to exactly rw, suid, dev, exec, auto, nouser and async - it does NOT include noatime. If you want noatime you must write it, as the / line in the fstab above does. Note also that the kernel applies relatime by default regardless, which is why mount often reports relatime on a filesystem whose fstab line says only `defaults`.

$ mount | grep -E 'ext4|xfs|btrfs'

/dev/sda1 on /boot type ext4 (rw,relatime,seclabel)
/dev/sda2 on / type ext4 (rw,noatime,seclabel)
/dev/sda3 on /var type ext4 (rw,noatime,seclabel)
/dev/sdb1 on /data type ext4 (rw,noatime,seclabel)

Option	Meaning	When to use
`defaults`	rw, suid, dev, exec, auto, nouser, async	Default baseline
`noatime`	Skip access-time updates	Read-heavy workloads (3-5% throughput win)
`relatime`	Update atime only if older than mtime/ctime	Compromise between correctness and performance
`errors=remount-ro`	Remount read-only on filesystem error	Safer failure mode for production
`noexec`	Forbid execution of binaries	/tmp, /var/tmp, removable media
`nodev`	Forbid device-file interpretation	Filesystems that should not contain device files
`nosuid`	Ignore setuid/setgid bits	Filesystems that should not honour privilege escalation
`_netdev`	Wait for network before mounting	NFS, iSCSI, network filesystems
`nofail`	Do not abort boot if this mount fails	Non-critical mounts

What actually mounts your fstab entries at boot

The kernel’s only involvement is mounting the root filesystem, from the root= kernel parameter, usually pivoted from the initramfs. It never opens /etc/fstab.

Everything after that is systemd:

systemd-fstab-generator runs before any unit starts. It reads /etc/fstab and writes a .mount unit per entry into /run/systemd/generator/, named after the mount point — /var/log becomes var-log.mount. Read the result with systemctl cat var-log.mount.
Dependencies come from the paths, not the pass column. var-log.mount gets RequiresMountsFor=/var, so nesting orders itself. _netdev adds ordering after network-online.target. The pass column does not order anything.
systemd-fsck-root.service checks the root filesystem and systemd-fsck@.service checks the others, instantiated per device. Pass 0 means no fsck unit is generated at all; 1 and 2 both produce one, with root handled separately.
Failure handling is the nofail option’s job. Without it the mount is required by local-fs.target, so a failure fails the target and the boot lands in emergency.target. With it, the boot continues without that filesystem.

The practical consequence: when a mount misbehaves at boot, the evidence is in the unit, not in the file.

systemctl --failed                     # did a .mount or .fsck unit fail?
systemctl status var-log.mount
journalctl -b -u systemd-fsck@dev-sdb1.service

fstab syntax - fields, options, and the dump/pass columns

The six columns

Common mount options

fstab discipline

Knowledge check