mount and findmnt

Concept

mount attaches a filesystem to a directory in the VFS tree. findmnt (util-linux) is the inspection tool: it reads the kernel mount table and can compare it to /etc/fstab. Use findmnt to see; use mount/umount (or a systemd .mount unit) to change.

A path is not a disk. The same directory name can sit on root, on a bind mount, on NFS, or on a container mount namespace. findmnt -T <path> is how you stop guessing.

Why it matters

• “Disk full” on /var/lib is a different problem if /var is its own filesystem
• Emergency mode after reboot is often one bad fstab line
• NFS/CIFS hangs, remount-ro events, and bind-mount surprises all show up here first
• Services can be “active” while the directory they write to is still the empty mount point on root

Never edit fstab and reboot to “see if it works”. Verify first.

Mental Model

Source  (block dev, UUID, LABEL, NFS, tmpfs, overlay)
    → mount(2) + options (rw/ro, noexec, _netdev, defaults)
        → target directory (must already exist)
            → visible in this mount namespace

Persistence:
  /etc/fstab          → mount -a / systemd-fstab-generator
  *.mount units       → After=/Requires= ordering with services
  runtime-only mounts → gone at reboot

findmnt without arguments prints the tree for the current namespace. Inside a container that tree is not the host’s.

Trailing detail that bites: bind mounts hide whatever was in the target directory. The files are not deleted; they are shadowed until umount.

Key Commands

# Inspection — prefer findmnt over parsing mount(8) output
findmnt
findmnt -D                          # df-like: source, size, used
findmnt -T /var/log                 # which mount owns this path?
findmnt -S /dev/sda1                # by source
findmnt -t xfs,ext4,nfs4
findmnt -o TARGET,SOURCE,FSTYPE,OPTIONS,AVAIL,USED
findmnt -R /mnt                     # recursive under a point (nested binds)
 
# Kernel table vs fstab
findmnt --fstab
findmnt --verify                    # catch bad UUIDs / missing targets before reboot
 
# Raw / classic
cat /proc/self/mountinfo            # namespace-accurate, richest kernel view
cat /proc/mounts
mount | column -t
 
# Mount by stable identifier (use these in fstab)
mount UUID=xxxxxxxx /mnt/data
mount LABEL=backup /mnt/backup
lsblk -f                            # UUID / LABEL / FSTYPE map
 
mount -a                            # all fstab entries except noauto
mount -o remount,ro /
mount -o remount,rw /
mount -o remount,noatime /var
 
# Unmount
umount /mnt/data
umount -l /mnt/data                 # lazy: detach now, release when last user leaves
umount -R /mnt                      # recursive (nested binds)
# umount -f                         # force; last resort on dead NFS, can lose writes
 
# Bind / rbind / move
mount --bind /src /dst
mount --rbind /src /dst             # include nested mounts under /src
mount --make-rprivate /dst          # stop propagation surprises (containers)
mount --move /old /new

Minimal fstab line worth copying:

UUID=...  /data  xfs  defaults,nofail,_netdev  0  2

nofail lets boot continue if the volume is missing. _netdev waits for the network. Neither is a substitute for a correct unit dependency.

Common Failure Modes & Symptoms

Symptom	Likely cause	First checks
Mount point empty / unexpected files	Not mounted, or bind shadowing the real tree	findmnt -T /path, lsblk -f
target is busy on umount	cwd, open FD, or nested mount	lsof +f -- /path, findmnt -R /path, fuser -vm /path
Emergency / rescue after reboot	Bad UUID, missing device, NFS without _netdev	findmnt --verify, comment the line, journalctl -b -u initrd-fs
NFS/CIFS mount hangs the shell	Hard mount + dead server	another tty; umount -l; see NFS note
Filesystem remounted read-only	Journal / barrier error	dmesg, findmnt -o OPTIONS /path, plan fsck
/ full, data volume shows free	Path is still on root	findmnt -T /var/lib/app, df -h /var/lib/app
Service writes vanished on reboot	Wrote into empty mount point before unit mounted	RequiresMountsFor=, .mount Before= the service
Options you set are missing	fstab defaults overrode a remount, or helper ignored flags	findmnt -o OPTIONS; remount explicitly

Investigation Tips

• findmnt -T then df -h on that exact path. Do not df the parent and assume.
• After any fstab edit: findmnt --verify && mount -a && findmnt before you leave the session, let alone reboot.
• Prefer UUID or LABEL. /dev/sdX and /dev/nvme0n1p2 reorder.
• Lazy unmount is the safe “get this path out of the tree” move. Force unmount on a network FS can discard uncommitted writes.
• systemd: systemctl status mnt-data.mount and systemctl list-dependencies --after <service> tell you whether the app waited for the volume.
• mountinfo fields include the mount ID and parent ID — that is how you see bind stacks and propagation.
• In containers, check mounts from the host with nsenter -t <pid> -m findmnt when the in-container view is incomplete.

Related Notes

• Filesystems and Mounts
• lsblk
• Block Devices and Partitions
• df and du Deep Dive
• NFS Troubleshooting
• Disk Full Runbook
• systemd Units
• Troubleshooting Methodology

Personal Lessons Learned

• An app “lost” a week of uploads because the NFS unit came up after the service. The process wrote into the empty local directory on root, filled /, and the real share mounted over the evidence. RequiresMountsFor=/data would have blocked the start.
• umount -f on a busy NFS export left the application with ESTALE on every open file. Lazy unmount plus restart of the writers was slower and actually finished.
• findmnt --verify caught a copied fstab UUID from another host. That check is cheaper than a 20-minute trip to the console.