[HV22.25] Santa’s Prophesy

December 29, 2022 1 minute read

Based on an old fairytale from Kobeliaky, Santa can provide more than presents. He can show you the future!

Solution

When I did this challenge, we believe it was broken since the training data was wrong. Most people solved this challenge by accident, so did I.

We got a website:

website

Not much to see in there, besides some strange CSV file appended to the image, which was the broken training data. Let’s enumerate a bit. I used dirb with the common.txt and found /upload. This looks promising.

upload function

The text and the loading looks like it tries to load a pytorch model (in jit.script) and run over some data. Since I didn’t understand the training data at all, my first step was to get a model in a format they expect and can use. This model should do nothing for now but simply return the input value. The code I ended up with looks like this:

import torch
import requests

uri = '77de8133-abbc-4b1c-8753-d76f5f549668.idocker.vuln.land'
uri = f"https://{uri}/upload"


class NeuralNetwork(torch.nn.Module):
    def forward(self, x):
        return x


model = NeuralNetwork()
jitModel = torch.jit.script(model)
torch.jit.save(jitModel, './25/model.pt')

files = [
    ('file', ('model.pt', open('./25/model.pt', 'rb'), 'application/octet-stream'))]
res = requests.post(url=uri,
                    files=files)

print(res.content)

Yes, this also implements the upload process and gives back the message from the webserver.

Since I knew other people solved it by accident without GPU training of the model, I poked around. I used the documentation from pytorch to learn about Tensors and this Module I was building, and decided to try to leak some information. I tried lot’s of functions inside forward(), but none of them worked.

After being hinted that it’s one command, I just threw different commands at it and torch.atan(x) was the first one that worked.

import torch
import requests

uri = '77de8133-abbc-4b1c-8753-d76f5f549668.idocker.vuln.land'
uri = f"https://{uri}/upload"


class NeuralNetwork(torch.nn.Module):
    def forward(self, x):
        return torch.atan(x)


model = NeuralNetwork()
jitModel = torch.jit.script(model)
torch.jit.save(jitModel, './25/model.pt')

files = [
    ('file', ('model.pt', open('./25/model.pt', 'rb'), 'application/octet-stream'))]
res = requests.post(url=uri,
                    files=files)

print(res.content)

This gave me the flag HV22{AA21B6AB-4520-4AD2-8016-4A9F2C371E6E}

Daniel Henschel

Solution