How much space a file takes up: du

Before you delete or archive something on a server, you need to know its size. How much the backup weighs, which directory ate the disk, whether the archive will fit on a flash drive: du (disk usage) answers these questions. The command shows how much space a file or directory takes up on disk.

Where to practice

The examples below work in an ubuntu:24.04 container. You can bring up the same one yourself:

docker run --rm -it ubuntu:24.04 bash

On Linux and in WSL2 the commands work right in your own terminal too. On macOS du is different (it comes from BSD); see "Common mistakes" for the details.

Create a file exactly 8 kilobytes long:

dd if=/dev/zero of=/tmp/big bs=1024 count=8 2>/dev/null

dd copies 8 blocks of 1024 bytes from /dev/zero (an infinite stream of zero bytes) into /tmp/big. 2>/dev/null hides dd's status report.

What du prints

du /tmp/big

8	/tmp/big

Two fields: size and file name, separated by a tab. Units are set with a flag:

• du -k in kilobytes, as a whole number. In GNU du this is already the default unit, but with -k the result is the same on any system.
• du -m in megabytes.
• du -h for humans: it picks K, M, or G on its own.

du -h /tmp/big

8.0K	/tmp/big

-h is easy on the eyes but awkward in a script: 8.0K can't be compared with a number. For a script, use -k or -m.

How to leave just one number

With wc from the last lesson, the trick with < helped. It doesn't work with du: du doesn't read stdin, it needs a file name. Without a name, it counts the current directory:

du -k < /tmp/big

12	.

So you strip the name afterward, from the output. The fields are separated by a tab, and cut -f1 cuts out the first field (by default cut splits a line on tabs):

du -k /tmp/big | cut -f1

8

awk '{print $1}' does the same thing: it prints the first field, splitting on spaces and tabs. You can save the number to a file with >, as in the last lesson.

Size on disk and file length

ls -l also shows a size, but a different one. It's the file's length in bytes, how much data is actually written to it. du shows how much space the file system allocated for the file. Space is allocated in blocks, usually 4 KB each, so even a tiny file takes up a whole block:

printf 'hello\n' > /tmp/small
ls -l /tmp/small

-rw-r--r-- 1 root root 6 Oct  9 18:27 /tmp/small

du -k /tmp/small

4	/tmp/small

6 bytes of data, 4 KB on disk. A file with 10,000 bytes would take up three blocks, and du -k would print 12. du can also show the file's length in kilobytes, with the --apparent-size flag (length, not space):

du -k --apparent-size /tmp/small

1	/tmp/small

It also happens the other way around: huge length, almost no space. That's a sparse file. The file system doesn't store blocks that were never written to. These files are often used for virtual machine disk images:

truncate -s 1G /tmp/sparse
ls -lh /tmp/sparse

-rw-r--r-- 1 root root 1.0G Oct  9 18:27 /tmp/sparse

du -h /tmp/sparse

0	/tmp/sparse

When disk space runs out, trust du: it shows what's actually in use.

Directory size

For a directory, du by default prints a line for every subdirectory and a grand total at the end. Usually you only need the total, which you get with -s (summary):

du -sh /var/log

To find which subdirectory is the heaviest, measure each one and sort:

du -sh /var/log/* | sort -h

sort -h understands the suffixes K, M, G, so 900K ends up above 2.0M. The biggest will be the last line. The asterisk * doesn't catch hidden directories (names starting with a dot). If they matter, use du -h -d 1 /var/log | sort -h: -d 1 means "one level deep", and hidden directories will be included too.

A directory's size in ls -l (4.0K for each) is the size of the directory entry itself, not of its contents. For directories, ls is useless: you need du.

Common mistakes

• du -k /tmp/big > /tmp/answer: the file will end up with 8 and the file name, separated by a tab. You need | cut -f1 or | awk '{print $1}'.
• du -k < /tmp/big: du doesn't read stdin and prints the size of the current directory.
• du -h in a script: 8.0K isn't a number. For comparisons, use -k.
• The size from ls -l instead of du: for small and sparse files the numbers can differ a lot. Dividing ls -l's output by 1024 is not the same as du -k.
• du -sh dir/* doesn't see hidden directories, and ls -lh dir doesn't show the size of a directory's contents.
• macOS: BSD du without a flag counts in 512-byte blocks, so an 8 KB file will show up as 16. With -k and -h the numbers match Linux. There's no --apparent-size flag there (there's -A), and -d 1 works.

What you'll need for the tasks

• Get a file's size with du -k and leave just one number with cut -f1 or awk.
• Save the number to a file with >.
• Compare directories with du -sh .../* | sort -h, and don't trust ls -l when it's about space used.