swapoff — disable devices and files for paging and swapping

Overview

swapoff deactivates a swap area: it tells the kernel to stop using the given device or file and to pull every swapped-out page back into RAM (or push it to another swap area). It is the counterpart of swapon and the standard preamble for touching a swap device — partition resizing, disk removal, filesystem work on the volume holding a swapfile. It ships with the mount package (Debian bookworm) at /usr/sbin/swapoff and requires root.

swapoff is often confused with swapon (activates; note the asymmetry: swapoff -a reads /proc/swaps, swapon -a reads /etc/fstab), with mkswap (writes the header, activates nothing), and with deleting the swapfile (which is only safe after a successful swapoff).

Synopsis

swapoff [options] [<spec>...]

Common one-line forms:

swapoff /dev/sda2                   # disable one partition
swapoff /swapfile                   # disable a swapfile
swapoff -a                          # disable everything in /proc/swaps
swapoff -v LABEL=swap1              # by label, with progress messages

How It Works

The drain-back operation

swapoff wraps the swapoff(2) syscall. The kernel must unswap: every page of the area currently in use is read back into memory — if RAM (plus remaining swap) cannot hold them, the syscall fails rather than losing data:

 swap area (inactive target)          RAM
 ┌─────────────────────────┐      ┌──────────────┐
 │ pages of process images │ ───► │ pages land   │
 │ marked "in swap on X"   │      │ in RAM /     │
 └─────────────────────────┘      │ other swap   │
        kernel clears the         └──────────────┘
        swap map for X, then
        removes X from /proc/swaps

This is why swapoff is slow on a busy system (it moves potentially gigabytes synchronously) and why it can fail with “Cannot allocate memory” on a loaded box — the classic operational trap.

Reading /proc/swaps like swapoff -a does

Each line of /proc/swaps is one active area; -a iterates exactly this list, so the file is both the source of truth and the progress indicator during a long drain-back (Used should fall toward 0):

Filename                Type       Size     Used  Priority
/dev/sda2               partition  4194300  1024  -2
/swapfile               file       2097148  0     100

Type distinguishes partitions from files, Priority reflects -p/ fstab pri= values (negative = kernel default), and an area disappears from the file the moment its swapoff(2) completes.

Specs: device, file, label, UUID

An explicit target may be given as a device path, a swapfile path, or a LABEL=/UUID= reference (also bare -L label / -U uuid forms) — resolved through blkid, mirroring fstab conventions. With -a it disables everything the kernel currently has active, enumerated from /proc/swaps — not from fstab.

$ cat /proc/swaps
Filename    Type    Size   Used   Priority
/dev/sda2   partition 4194300 0    -2

systemd-managed swap

On modern systems swap units are generated for fstab entries and device units appear for block devices. A manual swapoff succeeds, but the activation method matters for what happens next: a systemctl swapoff <unit>-style view (or editing fstab) is what prevents the area from coming back at the next activation pass or reboot. Simply swapoff-ing a device that a unit still references invites reactivation.

Options That Matter

The option surface is deliberately small — the complexity lives in the kernel’s drain-back, not in flags.

Usage Patterns

# Shrink or remove a swap partition: deactivate first (root)
swapoff /dev/sda2

# Disable everything currently active (reads /proc/swaps)
swapoff -a

# Remove a swapfile safely — deactivate, then delete
swapoff /swapfile && rm /swapfile

# By label or UUID, as fstab would reference it
swapoff -L swap1
swapoff UUID=7a3f1c2e-1111-4e5f-8a9b-001122334455

# Refresh all swap with new priorities from fstab
swapoff -a && swapon -a

# Watch progress on a large area (verbose; can take minutes)
swapoff -v /dev/sdb2

# Before resizing the volume that holds swap (LVM)
swapoff /dev/vg0/swap && lvreduce -r -L 2G /dev/vg0/swap && swapon /dev/vg0/swap

# Check what is (still) active afterwards
cat /proc/swaps; swapon --show

Nuances and Gotchas

• It can simply fail under memory pressure. Drain-back needs RAM (or other swap) for every resident swapped page; on a loaded system you get swapoff failed: Cannot allocate memory. Free memory first, reduce load, or lower the area’s priority so pages migrate to other swap.
• It can take a long time, synchronously. No progress bar, no timeout; a multi-GiB swapoff holds your shell. Run it in tmux/under nohup for big boxes and monitor /proc/swaps Used dropping.
• -a reads /proc/swaps, not fstab. swapoff -a disables what is actually on; swapon -a enables what should be on per fstab. The asymmetry is intentional and frequently mistaken.
• Root only. A non-root caller is refused up front with swapoff: Not superuser. (observed exit status 16 on recent util-linux) — an early check, distinct from a syscall failure.
• Deleting a swapfile without swapoff corrupts nothing immediately but leaves pages pointed at a missing file; the kernel holds the inode, so disk space is not reclaimed until swapoff — and swapon on the path later creates a fresh, unrelated area.
• systemd will bring swap back. fstab-generated .swap units and device units reactivate areas; persist changes by editing fstab (or masking units), not by swapoff alone.
• Hibernation images live in swap. Swapoff on the resume device destroys a pending hibernation image; expect a cold boot instead.
• Disabling an inactive area is an error. The kernel’s swapoff(2) returns EINVAL for a target that is not currently active, so retry loops should re-check /proc/swaps rather than blindly re-running the command.
• swapoff -a during performance testing is a legitimate move (show pure-RAM behavior) — but re-enable afterwards; a swapless box OOM-kills under pressure that swap would have absorbed.

Exit Status

The man page documents no code table; script against 0 vs non-zero and read stderr for the cause.

Related Commands

• swapon — the activating counterpart; note the fstab vs /proc/swaps asymmetry.
• mkswap — writes the header that makes an area swappable in the first place.
• swaplabel — edit the label/UUID used to reference the area.
• lsblk — identify swap partitions when planning removals.
• systemd — generated .swap units decide whether an area comes back.
• util-linux overview — collection hub for the other util-linux pages.

Interview Questions

Q: Why does swapoff fail with “Cannot allocate memory” on a loaded system?

swapoff must read every in-use swapped page back into RAM (or push it to another swap area). If MemAvailable plus other swap cannot hold them, the kernel refuses rather than evicting or losing data. Remedies: free memory (drop caches won’t help much — drop load instead), activate other swap first so pages have somewhere to go, or reduce the working set.

Q: swapoff -a ran on a system where /etc/fstab lists two swap areas but only one was active. What happened?

Exactly one area — the active one — was disabled: swapoff -a iterates /proc/swaps, the kernel’s live table, not fstab. The inactive fstab entry is untouched (and still configured for future swapon -a). This asymmetry with swapon -a (which reads fstab) is a favorite gotcha.

Q: You deleted an in-use swapfile with rm. What is the state of the system?

The kernel keeps the inode and its pages; /proc/swaps still shows the entry and df does not reclaim the space. The system works but the storage is leaked until swapoff completes on that area. Correct sequence is always swapoff /swapfile && rm /swapfile, then update fstab.

Q: How do you safely shrink the LVM logical volume that holds swap?

swapoff /dev/vg0/swap, perform the lvreduce (with -r to shrink the file layer if it is a swapfile on a filesystem — for a raw swap LV, just the LV), then re-run mkswap if the size changed in a way that invalidates the header (or rely on swapon’s checks), and swapon again. Touching the volume while swap is active risks kernel panics or I/O errors, which is why swapoff is the mandatory first step.

Q: What is the difference in outcome between swapoff -a and disabling the fstab entry?

swapoff -a changes only the running system: the next boot (or the next systemd activation pass) re-enables everything fstab still lists. Persisting the change requires editing fstab (or masking the generated .swap unit). Use both: swapoff for now, fstab edit for forever.

Q: A pending hibernation image exists and someone runs swapoff -a. Consequences?

The hibernation image is stored in swap; deactivating the area destroys it. The machine will cold-boot instead of resuming, losing the suspended session. Anything that manipulates swap on a laptop/server with hibernation enabled must check for a pending image first — one of the less obvious operational risks of swapoff.

References

• Man page(8) — manpages.debian.org
• Source — GitHub
• Source — Debian sources