I’ve always had a thing for old-school web tech. By the time I joined the digital fray, CGI scripts were pretty much relics, but the term kept popping up in tech forums and discussions like ghosts from the past. So, I got curious, started reading about them, and wanted to see if I could reason about them from the first principles. Writing one from the ground up with nothing but Go’s standard library seemed like a good idea.

Turns out, the basis of the technology is deceptively simple, but CGI scripts mostly went out of fashion because of their limitations around performance.

What are those

CGI scripts, or Common Gateway Interface scripts, emerged in the early 1990s as a solution for creating dynamic web content. They acted as intermediaries between the web server and external applications, allowing servers to process user input and return personalized content. This made them essential for adding interactivity to websites, such as form submissions and dynamic page updates.

The key function of CGI scripts was to handle data from web forms, process it, and then generate an appropriate response. The server then takes this response and displays it on a new web page. Here’s how the process might look:

sequenceDiagram participant U as Client participant S as Server participant C as CGI Script U->>S: Post request with a dynamic field value S->>C: Execute the CGI script in a new process Note right of C: CGI script receives the value C-->>S: Process and return result S-->>U: Respond with result

How to write one

CGI scripts are usually written in dynamic scripting languages like Perl, Ruby, Python, or even Bash. However, they can also be written in a static language where the server will need execute the compiled binary. For this demo, we’re going to write the server in Go using the cgi stdlib, but the CGI script itself will be written in Bash.

Here’s the plan:

  • Set up a basic HTTP server in Go.
  • The server will await an HTTP POST request containing a form field called name.
  • Upon receiving the request, the server will extract the value of name.
  • Next, it’ll set the $name environment variable for the current process.
  • A Bash CGI script is invoked, which uses the $name environment variable to echo an HTML-formatted dynamic message.
  • Finally, the server will then return this HTML response to the client.

The server lives in a single main.go script. I’m leaving out Go’s verbose error handling for clarity.

package main

import (
    "net/http"
    "net/http/cgi"
    "os/exec"
)

// Leaves out error handling for clarity
func cgiHandler(w http.ResponseWriter, r *http.Request) {
    // Parse name from post request
    r.ParseForm()
    name := r.FormValue("name")

    // Execute the CGI script with the name as an environment variable
    cmd := exec.Command("cgi-script.sh")
    cmd.Env = append(cmd.Env, "name="+name)

    // Serve the CGI script
    handler := cgi.Handler{Path: cmd.Path, Dir: cmd.Dir, Env: cmd.Env}
    handler.ServeHTTP(w, r)
}

func main() {
    http.HandleFunc("/", cgiHandler)
    http.ListenAndServe("localhost:8080", nil)
}

Upon every new request, the server above will execute a CGI script written in Bash. Name the shell script as cgi-script.sh and place it in the same directory as the server’s main.go file. Here’s how it looks:

#!/bin/bash

set -euo pipefail

name=$name

echo "Content-type: text/html"
echo ""
echo '<html><body>'
echo "Hello $name, greetings from bash!"
echo '</body></html>'

The script just reads name from the environment variable, sets the Content-Type header, injects the value of name into the message, and echos the out the final HTML response. The server then just relays it back to the client. To test this:

  • Run the server with go run main.go.
  • Set the permission of the CGI script:
    sudo chmod +x cgi-script.sh
  • Make a cURL request:
    curl -X POST http://localhost:8080 -d "name=Redowan"

This returns the following response:

<html><body>
Hello Redowan, greetings from bash!
</body></html>

Why they didn’t catch on

CGI scripts have fallen out of favor primarily due to concerns related to performance and security. When a CGI script is executed, it initiates a new process for each request. While this approach is straightforward, it becomes increasingly inefficient as web traffic volume grows. However, it’s worth noting that modern Linux kernels have made improvements in process spawning, and solutions like FastCGI utilize persistent process pools to reduce the overhead of creating new processes. Nevertheless, you still incur the VM startup cost for each request when using interpreted languages like Python or Ruby.

Modern application servers like Uvicorn, Gunicorn, Puma, Unicorn or even Go’s standard server have addressed these inefficiencies by maintaining persistent server processes. This, along with the advantage of not having to bear the VM startup cost, has led people to opt for these alternatives.

Another concern worth considering is the evident security issues associated with CGI scripts. Even in our simple example, the Bash script accepts any value for the name parameter and passes it directly to the response. This exposes a significant vulnerability to injection attacks. While it’s possible to manually sanitize the input before passing it to the next step, many of these security steps are automatically handled for you by almost any modern web framework.

Fin!

  1. Apache Tutorial: Dynamic Content with CGI1 ↩︎

Recent posts