Newline and CRLF command injection
Newline command injection uses a line-feed (`%0a`) or carriage-return/line-feed (`%0d%0a`) character to end one command and begin another. A raw newline is a command terminator in the shell just like a semicolon, so when `;`, `&&` and `|` are filtered, an injected newline — often URL-encoded through an HTTP parameter — still splits the command.
How it works
In a shell script or a `sh -c` string, a newline terminates the current command. If user input containing a literal LF is concatenated into a command, everything after the newline is parsed as a new command line. Delivered over HTTP, the newline is usually percent-encoded as `%0a` (LF) or `%0d%0a` (CRLF) and decoded into the command string by the application.
This is especially relevant where input is written into a config file, a crontab, or a command executed line-by-line, and where filters target the obvious operators but forget that a newline is itself a separator. It overlaps with CRLF injection in other contexts, but here the sink is a command interpreter.
Representative payloads
Illustrative fragments for authorized testing. The injected marker commands (id, whoami) are harmless proofs of execution — swap in your engagement’s agreed markers.
%0aidLinux / UnixURL-encoded LF terminates the line; id runs on the next line. Bypasses ;/&&/| filters.
%0d%0aidLinux / UnixWindowsCRLF variant for sinks that expect Windows-style line endings.
value\nidLinux / UnixLiteral newline (shown as \n) injected where the application decodes escape sequences or accepts multiline input.
%0a/bin/id%0aLinux / UnixNewlines on both sides isolate the injected command on its own line to avoid syntax errors.
When it works
- User input (often from an HTTP parameter) is decoded and concatenated into a shell command or a line-oriented file that gets executed.
- Other separators are filtered but newline/CRLF characters are not.
When it fails
- No shell/line-oriented interpreter processes the input.
- Newlines are stripped or rejected, or the value is allowlisted.
Detection
Flag encoded and literal newlines in command-driving inputs and watch for multi-line command strings reaching exec.
Signals to look for
- Parameters containing %0a, %0d, or literal CR/LF where single-line values are expected.
- Command strings or config/cron entries that gain extra lines after a request.
- Process telemetry showing a second command that corresponds to post-newline content.
Where to look
Remediation
The correct fix: Execute shell-free with an argv array and reject CR/LF in single-line values; a newline is then just a data byte.
Principles
- Shell-free execution treats embedded newlines as literal argument content, not a new command.
- Strip or reject CR/LF in inputs that must be single-line (hostnames, filenames, IDs).
- When writing user input into config/cron files, encode it and never let it introduce new directives.
Safe APIs by runtime
execFile/spawn with args array; validate values are single-line.
if (/[\r\n]/.test(v)) reject(); execFile("host", ["--", v]);subprocess.run([...], shell=False); reject CR/LF in single-line fields.
if any(c in v for c in "\r\n"): abort()Not a fix: Filtering only ';' and '&&'. A newline is an equally valid command terminator and is easy to deliver as %0a over HTTP.
For the full secure-coding matrix across languages, see the command injection prevention guide.