Hands-on Lab·CompTIA Linux+

Swap: a file, a partition, and how eagerly it is used

Domain 1 names swap management. This guide reads what a machine already has, adds a swap file by hand with the permission that matters, adds a second area with a priority so you can see which one wins, changes how eagerly the kernel pages, and removes everything - with the numbers that say whether swap is being used at all.

System Management Guide 18 of 28 Beginner

OSUbuntu 26.04 LTS
Kernel7.0.0-30-generic
systemd259
Mandatory access controlAppArmor on Ubuntu, SELinux enforcing on Alma
TimeAbout 22 min

Firewallufw / nftables 1.1.6 - firewalld / nftables 1.1.5

One host. Swap works identically on the RPM family; only the file that makes a setting persist differs, and the page says where it lives.
Server Name	IP Address	OS	Roles	CPU	RAM	HDD
LPLUS-A01	192.168.0.73	Ubuntu 26.04 LTS	Debian-family host - apt, ufw, netplan, AppArmor	2 Core	4 GB	50 GB

This guide includes

Use this when a machine needs more swap without a reboot, or when someone proposes changing swappiness to fix a performance problem. This matters because a swap file created the obvious way is world-readable until you fix it, and because swappiness is a weighting rather than a percentage.

reading the existing swap areas three ways, and the swappiness that governs them
creating a swap file with the right permission before it is activated
adding a second area and setting the priority that decides which is used first
changing swappiness in the running kernel and knowing where it persists
reading whether the machine is actually paging, rather than assuming

Before you start

lvm-from-physical-volume-to-a-filesystem-that-grows

What the machine has now
Three views of the same thing, and the number that governs it.
bash Example session
```
$free -h | head -3; echo "---"; swapon --show; echo "---"; cat /proc/swaps               total        used        free      shared  buff/cache   availableMem:           3.3Gi       548Mi       1.7Gi       7.9Mi       1.4Gi       2.8GiSwap:          3.8Gi          0B       3.8Gi---NAME      TYPE SIZE USED PRIO/swap.img file 3.8G   0B   -1---Filename				Type		Size		Used		Priority/swap.img                               file		4003836		0		-1$cat /proc/sys/vm/swappiness; sysctl vm.swappiness; echo "--- the same number, two spellings"60vm.swappiness = 60--- the same number, two spellings
```
Note
swapon --show is the readable form and /proc/swaps is the raw one. The second reports sizes in 1 KiB blocks, which is why the numbers look unfamiliar: 4003836 blocks is the same 3.8 GiB.

Ubuntu ships a swap FILE, not a partition. /swap.img is a file on the root filesystem, which is the modern default because a file can be resized and a partition cannot. The type column says which you have.

Priority -1 means the kernel chooses. Areas with equal priority are used round-robin; a higher number is preferred. The next step sets one deliberately.

Swappiness 60 is the default and is not a percentage. It is a weighting between reclaiming page cache and swapping anonymous memory, on a scale of 0 to 200. Setting it to 0 does not disable swap - it makes the kernel avoid it until it has almost no choice.

Expected result3.8Gi of swap, none used - swapon --show naming /swap.img as type file with priority -1, /proc/swaps saying the same in blocks - and a swappiness of 60.

Success conditionYou can describe a machine's swap configuration before changing it.
A swap file, made by hand
Four steps, and the one that must come before the others.
bash Example session
```
$sudo fallocate -l 256M /var/tmp/swapfile && sudo ls -lh /var/tmp/swapfile | awk '{print $1, $5, $9}'-rw-r--r-- 256M /var/tmp/swapfile$sudo chmod 600 /var/tmp/swapfile && sudo mkswap /var/tmp/swapfile 2>&1 | tail -2; sudo ls -l /var/tmp/swapfile | awk '{print $1, $9}'Setting up swapspace version 1, size = 256 MiB (268431360 bytes)no label, UUID=366425a0-a049-4472-8e14-d3ca538144c5-rw------- /var/tmp/swapfile$sudo swapon /var/tmp/swapfile && swapon --show; free -h | grep -i swapNAME              TYPE SIZE USED PRIO/swap.img         file 3.8G   0B   -1/var/tmp/swapfile file 256M   0B   -1Swap:          4.1Gi          0B       4.1Gi
```
Note
fallocate created it mode 644 and that is the problem. A swap file contains whatever was in memory - keys, passwords, anything - so it must be readable only by root. chmod 600 BEFORE mkswap and swapon; modern swapon warns about insecure permissions, and a warning is easy to ignore.

Four steps in order: allocate, permission, format, activate. fallocate is instant; dd if=/dev/zero is the older way and is needed on filesystems where fallocate produces a sparse file, which swap cannot use.

mkswap writes a header with a UUID, which is what lets fstab reference it by UUID= rather than by path - the same argument as any other filesystem.

To make it permanent, add a line to /etc/fstab: /var/tmp/swapfile none swap sw 0 0. Without that the area is gone at the next boot, which is a common half-finished task.

Expected resultA 256M file created -rw-r--r--, then -rw------- after the chmod, mkswap reporting swapspace version 1 with a UUID - and swap totalling 4.1Gi once it is on.

Success conditionYou can add swap to a running machine without a reboot.
A second area, and which one is used first
Priority, made visible.
bash Example session
```
$sudo truncate -s 128M /var/tmp/swap.img && D=$(sudo losetup -f --show /var/tmp/swap.img) && echo $D | sudo tee /tmp/swap.dev && sudo mkswap $D > /dev/null && sudo swapon -p 10 $D && swapon --show/dev/loop0NAME              TYPE      SIZE USED PRIO/swap.img         file      3.8G   0B   -1/var/tmp/swapfile file      256M   0B   -1/dev/loop0        partition 128M   0B   10$cat /proc/swaps; echo "--- higher priority is used first, and -1 means the kernel decides"Filename				Type		Size		Used		Priority/swap.img                               file		4003836		0		-1/var/tmp/swapfile                       file		262140		0		-1/dev/loop0                              partition	131068		0		10--- higher priority is used first, and -1 means the kernel decides
```
Note
Higher priority is used first. With -p 10 the loop device outranks both files, so it fills before they are touched. That is how you put swap on fast storage and keep a slower area as overflow.

Equal priorities are used round-robin, striping across areas, which is worth having when they are on separate disks.

The type says partition for a loop device, not file. The column describes the backing device rather than how you made it - a loop device IS a block device to the kernel, which is the same reason the LVM page can be followed without a spare disk.

In fstab the priority is a mount option: sw,pri=10. Setting it with swapon -p is runtime only, like everything else on this page.

Expected resultThree areas: /swap.img and /var/tmp/swapfile at priority -1, and /dev/loop0 at 10 - and its type reported as partition.

Success conditionYou can control which swap area the kernel reaches for first.
How eagerly the kernel swaps
One number, changed twice, and where it would persist.
bash Example session
```
$sudo sysctl -w vm.swappiness=10; cat /proc/sys/vm/swappiness; echo "--- changed in the running kernel only"vm.swappiness = 1010--- changed in the running kernel only$ls /etc/sysctl.d/ | head -4; echo "--- and a file here is what makes it survive a reboot"README.sysctl--- and a file here is what makes it survive a reboot$sudo sysctl -w vm.swappiness=60 > /dev/null; sysctl vm.swappiness; echo "--- put back to the default"vm.swappiness = 60--- put back to the default
```
Note
sysctl -w writes the running kernel only. It is exactly equivalent to echoing into /proc/sys/vm/swappiness, and both are lost at the next boot.

Persistence is a file in /etc/sysctl.d/ ending in .conf, then sysctl --system to load every file in order. This machine has none - only the README - so every sysctl on it is currently at its default.

Lowering swappiness does not stop swapping. At 10 the kernel prefers to drop page cache first, but it will still swap rather than fail an allocation. Only swapoff removes swap from the picture, and that is a different decision with different consequences.

Databases often want a low value and general-purpose servers rarely need any change at all. Changing it to fix a performance problem, without first checking whether the machine is paging, is the mistake the next step is about.

Expected resultvm.swappiness = 10 and the proc file agreeing - /etc/sysctl.d/ holding only README.sysctl - then back to 60.

Success conditionYou can change swappiness and say what would make it survive a reboot.
What swap is actually for
The numbers that say whether any of this matters.
bash Example session
```
$free -m | head -2; awk '/^(MemTotal|SwapTotal|SwapFree|Committed_AS|CommitLimit):/ {printf "%-14s %8.0f MB\n", $1, $2/1024}' /proc/meminfo               total        used        free      shared  buff/cache   availableMem:            3398         534        1715           7        1400        2864MemTotal:          3399 MBSwapTotal:         4294 MBSwapFree:          4294 MBCommitLimit:       5993 MBCommitted_AS:       361 MB$vmstat 1 2 | tail -1 | awk '{print "si (swap in):", $7, " so (swap out):", $8}'; echo "--- zero on both means nothing is being paged right now"si (swap in): 0  so (swap out): 0--- zero on both means nothing is being paged right now
```
Note
si and so are the answer. Swap in and swap out, in KB per second. Both zero means nothing is being paged, and a machine with gigabytes of swap ALLOCATED and none in use is entirely healthy - the used column in free says the same thing.

Swap being used is not itself a problem. The kernel pages out memory that has not been touched for a long time to make room for cache, which is a good trade. Sustained non-zero si/so is the symptom worth chasing.

CommitLimit and Committed_AS are the forward-looking pair: what the kernel is prepared to promise against what it has already promised. The limit here is larger than RAM because swap counts toward it.

A machine with no swap fails allocations instead of slowing down, and the OOM killer arrives sooner. That is a legitimate choice for a container host and a poor one for a general-purpose server.

Expected result3,399 MB of memory against 4,294 MB of swap, a CommitLimit of 5,993 MB with only 361 MB committed - and si 0, so 0 from vmstat.

Success conditionYou can say whether a machine is paging before changing anything about swap.
Putting the machine back
Both areas removed, and the original left alone.
bash Example session
```
$sudo swapoff /var/tmp/swapfile && sudo swapoff $(cat /tmp/swap.dev) && swapon --show; echo "--- exit $? : no swap areas of ours left"NAME      TYPE SIZE USED PRIO/swap.img file 3.8G   0B   -1--- exit 0 : no swap areas of ours left$sudo losetup -d $(cat /tmp/swap.dev); sudo rm -f /var/tmp/swapfile /var/tmp/swap.img /tmp/swap.dev; cat /proc/swaps; losetup -a | wc -l; echo "loop devices left"; free -h | grep -i swapFilename				Type		Size		Used		Priority/swap.img                               file		4003836		0		-10loop devices leftSwap:          3.8Gi          0B       3.8Gi
```
Note
swapoff must succeed before the file is deleted. It copies anything still paged out back into memory, so it can take time and can fail if there is not enough free RAM to take it - which is why removing swap from a busy machine is not instant.

Order again: swapoff, then detach, then delete. Deleting a file that is still active leaves the kernel using an unlinked inode, which is invisible and survives until reboot - the same deleted-but-open problem the troubleshooting guide on this path covers.

The machine's original /swap.img was never touched, which the final free confirms at 3.8Gi.

Expected resultOnly /swap.img left in swapon --show and /proc/swaps, 0 loop devices, and swap back to 3.8Gi.

Success conditionThe machine has exactly the swap it started with.

Troubleshooting

swapon warns that a swap file has insecure permissions.

Why: The file was created 644. A swap file holds the contents of memory, so anyone able to read it can read whatever was paged out.

Fix:chmod 600 <file> before mkswap and swapon. If it is already active, swapoff, fix the mode, and swapon again.
Swap added by hand is gone after a reboot.

Why: swapon is a runtime action. Nothing records it.

Fix:Add a line to /etc/fstab: /path/to/swapfile none swap sw 0 0, or UUID=<uuid> none swap sw,pri=10 0 0 for a device. Test with swapoff -a && swapon -a before trusting it to a reboot.
Swap is in use and the machine is assumed to be short of memory.

Why: Allocated swap and active paging are different things. The kernel pages out long-idle memory as a matter of course.

Fix:vmstat 1 and read si and so. Sustained non-zero values mean real paging; zeros with swap used mean the kernel tidied up earlier and nothing is wrong.
swapoff hangs or fails.

Why: Everything paged out must come back into RAM, and there may not be room.

Fix:Free memory first, or swap off one area at a time rather than swapoff -a. On a machine under pressure, plan the removal rather than issuing it interactively.

Official sources

← All Linux+ guides

Swap: a file, a partition, and how eagerly it is used

This guide includes

Before you start

What the machine has now

A swap file, made by hand

A second area, and which one is used first

How eagerly the kernel swaps

What swap is actually for

Putting the machine back

Troubleshooting

Official sources