csplit Command in Linux: Split Files by Pattern or Line Number

Use csplit when a text file has repeated section headers — log blocks, report sections, or XML-like markers — and you want each block in its own file.

Create a sample file and split on SECTION:

mkdir -p /tmp/csplit-lab && cd /tmp/csplit-lab
cat > report.txt << 'EOF'
HEADER LINE
line one
line two
SECTION
data a
data b
SECTION
data c
FOOTER
end
EOF
csplit report.txt '/SECTION/' '{*}'

Sample output (byte counts on stdout):

30
22
26

List the pieces:

ls -l xx*

Sample output:

-rw-r--r-- 1 root root 30 Jul  1 17:51 xx00
-rw-r--r-- 1 root root 22 Jul  1 17:51 xx01
-rw-r--r-- 1 root root 26 Jul  1 17:51 xx02

xx00 holds lines before the first SECTION; later files hold content starting at each match through the next split.

When you know exact line boundaries — for example "first 3 lines are header" — pass line numbers instead of regex patterns.

Run the command:

cd /tmp/csplit-lab
rm -f xx*
csplit report.txt 3 6

Sample output:

21
24
33

Inspect the first piece:

cat xx00

Sample output:

HEADER LINE
line one
line two

xx01 contains lines 4–6; xx02 is the remainder of the file.

Default xx00 names are fine for quick work. In scripts, -f sets the prefix and -b sets a sprintf-style suffix.

Run the command:

cd /tmp/csplit-lab
rm -f part*
csplit -f part -b '%02d.txt' report.txt '/SECTION/' '{*}'

Sample output:

30
22
26

List results:

ls part*

Sample output:

part00.txt
part01.txt
part02.txt

Sorted names make reassembly and downstream tools easier.

By default, the matching SECTION line stays in the output piece. --suppress-matched removes those delimiter lines from every file.

Run the command:

cd /tmp/csplit-lab
rm -f xx*
csplit --suppress-matched report.txt '/SECTION/' '{*}'
cat xx01

Sample output from xx01:

data a
data b

Compare with the plain split where xx01 still began with the SECTION line. Use this when markers are not part of the data you need.

Some patterns produce empty sections. -z deletes zero-length output files so you do not chase ghost pieces.

Run the command:

cd /tmp/csplit-lab
rm -f xx*
csplit -z report.txt '/NOMATCH/' '{*}'
ls xx*

Sample output:

78
xx00

Only the full-file piece remains because /NOMATCH/ never fired to create extra empty segments.

{*} splits at every match. {1} repeats the pattern once more after the first split — useful when you want a header file plus one body chunk before the rest merges.

Run the command:

cd /tmp/csplit-lab
rm -f xx*
csplit report.txt '/SECTION/' '{1}'
ls -l xx*

Sample output:

30
22
26
-rw-r--r-- 1 root root 30 Jul  1 17:51 xx00
-rw-r--r-- 1 root root 22 Jul  1 17:51 xx01
-rw-r--r-- 1 root root 26 Jul  1 17:51 xx02

With two SECTION lines in the file, {1} performs one additional split after the first — yielding three pieces total. Use a single line number when you only need exactly two parts.

On error, csplit normally removes output files. -k keeps whatever was already written — helpful in long pipelines when you still want partial results.

Run the command:

cd /tmp/csplit-lab
rm -f xx*
csplit -k report.txt 99999

Sample output:

78
csplit: '99999': line number out of range

Despite the error, the first piece remains:

ls xx*
cat xx00

Sample output:

xx00
HEADER LINE
line one
line two
SECTION
data a
data b
SECTION
data c
FOOTER
end

Line 99999 is past EOF, so everything landed in xx00.

csplit does not provide a merge command. Concatenate pieces in sorted name order to rebuild the original bytes (when nothing was suppressed).

Run the command:

cd /tmp/csplit-lab
rm -f xx*
csplit -q report.txt '/SECTION/' '{*}'
cat xx00 xx01 xx02 > restored.txt
diff -u report.txt restored.txt

Sample output:

(no diff output — files match)

Use the same prefix order you created (part00.txt, part01.txt, …) when custom -f / -b names were used.