Kirk Jackson's Page of Words - OWASP

Recent talks - I know what you did last summer

Kirk Jackson — Wed, 09 Mar 2011 08:17:00 GMT

This month I gave a similar talk to two user groups. The OWASP Wellington (and Auckland over video conference), and the Wellington .NET user group both invited me to speak on: "I know what you did last summer; The latest from the world of web hacks".

This was a fun talk to deliver. The focus was on recent web 'hacks' that had occurred in the past few months (I used a pretty general definition of 'hack'), but the main discussion was around the lessons that we could learn from these issues and what we could draw back into our own projects.

I think this talk had the most amount of interaction out of any of my previous talks. There was lively discussion about what the root cause of the problem was, whether it was even fixable at all, and we lamented the effects of 'users' :)

Since the .NET talk was a superset of the OWASP one (it was longer), I've included those slides below:

2011-03-09-WellingtonNet.pdf (2.07 MB)

Thanks for coming!

Kirk

Dean Carter: Ramblings of an ex-QSA

Kirk Jackson — Thu, 15 Jul 2010 04:47:41 GMT

Used to be a QSA (Qualified Security Assessor). There are now 8 in NZ.

The QSA wears the risk and signs you off for PCI compliance.

There are no silver bullets for PCI stuff.

"It's a hell of a roller-coaster ride"

He has seen 2.5 million credit card numbers in NZ, in the clear, in many website databases.

One guy Albert Gonzalez compromised 170 million credit cards across many large corporations.

PCI requirements:

"Protect stored data": 79% of orgs fail on this.

PAN (account data) must be unreadable when stored.

You can never store mag stripe data.

"Track and monitor all access to network resources and cardholder data"

"Develop and maintain secure systems and applications" - 56% of organisations fail on this

Rant:

1. Card holder data gets everywhere

2. Keep test and development environments out of scope. Don't use real live data in them.

3. The good: payment gateways and companies that handle cards - they do a good job. They outsource to experts.

The bad: small merchants with a few transactions. Cheap website with cheap hosting. Easily compromised.

The ugly: corporates. Great staff but don't make any progress.

If you're a merchant: find a compliant service provider.

4. If your a service provider: code well, make a noise about it. Make your solutions easy to assess for compliance. Keep in touch with your acquiring bank.

5. You need to evolve your security to address risks. You are allowed to exceed PCI standards.

6. New VISA best practices: you don't need to store the PAN any more, rely on your service provider to do it.

7. Do it properly, or don't use credit cards. Support your developers and give them training.

8. Storage of card data: Challenge it - why does the business need it? Get rid of old cards if you don't need them.

9. Checkbox security - don't just check the boxes. Exceed them.

10. OWASP top 10 - adopted by PCI DSS.

Two most useful links:

www.pcisecuritystandards.org

www.owasp.org

Parting thoughts:

- Use OWASP as a tool

- Don't confuse compliance and standards with security

- Chop up your credit cards!

Questions:

Why did you give up being a QSA?

It was really stressful

When collecting info and passing it on to a payment gateway, do you require an audit?

Different QSAs treat it differently. He believes the webserver is in scope if it's taking the card data. New version of standard coming out in October that may address in-memory stuff.

Why stop using credit cards? At least you get protection, unlike if you use debit cards?

Dean uses a low-value debit card.

How does PCI deal with it if you're using third-party libraries?

Payment application DSS will kick in if you're using it to resell.

Quintin Russ / Mike Jager - Hosting and Security

Kirk Jackson — Thu, 15 Jul 2010 04:24:29 GMT

Hosting and Web Apps
The Obscurity of Security

Quintin from SiteHost and Mike from Web Drive cover horror stories they've uncovered in website code when they've been rung up to fix something.

Security used to be the domain of systems admins and hosters, but developers have added more fancy features.

Website owners and developer blame their hosters when their sites are defaced.

What if security isn't part of the spec?

Make it part of the spec.

(Shift jobs if management won't let you make it part of the spec.)

Security starts early: Planning and design phase

- Research, talk to security people
- Get your team some security experience
- Reduce the attack surface
- Keep it simple: Don't build a CMS for a 5 page site
- Don't have an admin area, or use defense in depth to protect it

Not all apps are equal:

- Sometimes buying is better than building
- Everything has security holes
- Pick something good
- How does vendor approach security?
- Check the apps security history:
- If there are no holes, beware. If there are silly problems, beware.

RTFM:

- Read the OWASP top 10
- Read the OWASP books
- Read the install documentation and follow the "After installation" docs.
- e.g. Think about what you do when you unserialise stuff; don't trust untrusted user data

Development:

- Attack surface reduction
- Validate all your input
- Use source control, and know how it works.
- Watch out for rolling .svn, .git, .cvs directories: might show directory lists, source code, usernames
- svn checkout is an invalid installation method
- Look at all the files that are there! Especially free / open source apps you download

Data management:

- If you don't need it, don't store it
- If you need to keep it, how do you need to access it?
- Hash (with a salt), don't encrypt
- Keep production and development seperate
- Keep tabs on your data - size, growth rates, is data used by the code? Get rid of it.

Password strategy:

- Don't reuse credentials
- Weak usernames and passwords for db - common to see dbname = username = password
- Watch out for old staff members and old passwords

Filesystem security:

- Watch out for apps that use /tmp and friends, or require special directory permissions
- Learn how to chmod correctly. x is good enough for directory traversal.
- Watch out for log files in web root
- Beware test files eg phpinfo
- Don't leave old crap on your filesystem: Session files, template caches, zip files

Deployment:

- Automate as much as possible
- Don't blindly follow installation instructions
- Read them when you select the software, and understand what it's doing
- Don't use hosting control panels if you don't need them - they have high level access to the underlying system, and greatly increase your attack surface
- Use SSL for the content not just for the login pages
- Keep your websites separate - different trust level = different credentials

Backups:

- Keep your own backups - don't trust the providers ones. They protect from a catastrophic failure, and you could lose 12-24 hours of data
- Test them before you need to use them

Clouds:

- Don't ever use remote includes - including some third party code in your app!
- Minimise remote resource usage:
- How does your site react if the remote resource is gone?
- Take your own copy of AJAX libraries
- Do you need third party analytics for everything?
- Outsourcing data storage: What data are you uploading? Where is it hosted? Is it safe? Who has access to it? How are backups stored, and how long are they retained?

Software lifecycle management:

- Have a process for decommissioning, make sure you delete data and files that aren't used
- Make sure software is up to date
- Who monitors upstream releases? How quickly do you make patches? Who makes the call?

Monitoring:

- Monitor changes to your website content and uptime
- Check external access. Has your whitelist stopped working?
- DNS: Remember that DNS is an external dependancy. Has your domain been hijacked?

Politics:

- Make security a part of job description - managers and developers need to make security a priority and make it part of KPIs
- Get buy-in from non-technical staff

Talk to your hosting providers:

Talk to their security guys well in advance. Make sure your specific requirements are getting through to the technician who is doing the work (don't trust the salesperson).

Remember: It's your job to make sure it's working

Questions:

Including KPIs is a good thing, but you need to give developers the time to learn.

Graeme Neilson / Kirk Jackson: Tales from the Crypt0

Kirk Jackson — Thu, 15 Jul 2010 03:25:18 GMT

Thanks to everyone who came along to our talk at OWASP NZ Day 2010 today.

Also, a big thanks to the Wellington .NET user group crowd that came last night to listen to our practice run -- you'll be pleased to know that we dropped the discussion of hash extension attacks :)

Here are the slides for your downloading pleasure: tales-of-the-crypto.ppt (3.79 MB)

Metlstorm: Low Scuttling Chillicrab

Kirk Jackson — Thu, 15 Jul 2010 01:48:11 GMT

Adam is one of the organisers of Kiwicon, and has presented on this topic in Singapore.

Using tools to capture / probe network traffic.

If you compare to app/data recon tools like Maltego, network recon tools aren't as start of the art.

But... if you own the networks under this new fangled cloud stuff, then you own the whole environment.

It's hard to map out, search and investigate >= Class A

At the moment, only big countries can do that sort of investigation. Apparently countries are gearing up for 'Cyber Wars'.

But, individuals and corporates can get involved in the same activities of cyber-war or cyber-terrorism.

Scanning, pinging and trying exploits doesn't scale well - you have to do a lot of work and get lots of false hits.

You might get owned randomly - it's cheap to own more targets, and then figure out what to do with it later.

Targeting:

It's hard to target large numbers of IP addresses. The current tools can't scale to those kinds of numbers (and the pay services will get really expensive).

lowscuttlingchillicrab.com

So he built a geo-targeted network recon data acquisition system with a web interface, and scanned all of NZ and Singapore for conferences.

An interface to search over data.

"This is a highly secure router, stay away" - the open telnet port tells us so.

Cool things it does:

Searches over certificates
Screen captures remote desktop screens
Good for targeting: finding particular applications / devices / protocols
Good at finding other assets owned by a company outside of their own netblock
Helps us understand how many vulnerable things are sitting out there

The internals of the tool:

Version 1 was just to see how plausible it was to scan large chunks of the internet. Used lots of glued together tools like nmap etc.

Version 2 is now a simple python script that has been optimised for acquiring the data by scanning a whole country block over certain ports.

A few billion rows of data - use MongoDB to store data. Erlang, RabbitMQ, Python, Celery MQ, Python / Django frontend, GridFS distributed filestore.

Target selection:

How do you define what a country is? Is it domain names ending in .nz? Netblocks announced at peering exchanges? Address registry allocations? GeoIP?

He chose GeoIP as it simplified things - but misses out on .nz stuff hosted overseas.

Acquiring data:

Custom-tuned protocols to limit rates, fire up application to capture details for different protocols.

About 1.4B rows per complete scan of NZ and Singapore.

Need to optimise for search / retrieval as that's the primary use once the data is acquired.

Data mining:

Look for old boxes, boxes with self-signed certs, certain switches, domains etc.

Singapore: 377k boxes that talk HTTP - more than the number of live systems. 14k cisco boxes. 12k open RDP (one with background of Commonwealth Bank of Australia :))

IDS Avoidance:

He's not actually carrying out any intrusions. Only collecting banners, and complying with what they say.

IDSs don't necessarily detect them - only 7 complaints to ISP in NZ, and one funny one in Singapore.

People are watching - DNS PTR backscatter gives an idea of people watching and resolving domain names for IP address.

Portscans aren't very interesting these days. People notice, but don't do anything.

But not good for:

If you notice mis-configured systems, it's hard to do anything about it.

Giving it as public / bad guy access would be difficult and cause problems.

What about Shodan?

Scan whole world for 4 ports (21, 22, 23, 80), but not as many hosts or depth of coverage in NZ.

Sells commercial access to exported data.

What does it mean?

A search engine over this data makes it very powerful.

It's not that hard to do this sort of thing. It's probably already being done by military or crime industries. Cheap compared to a drug submarine :)

Questions:

What did the abuse mails say?

One from a Uni, two or three from an ISP and they noticed scanning of the SIP voice customers. A few of ZoneAlarm type people noticing.

Scanning boxes: Where were they hosted? Bandwidth out?

Domestically peered, gigabit to APE. It's not really bandwidth constrained, it's constrained by politeness. Turned off state tracking for outbound connections. Could probably do the whole country in 2 hours if you cranked it up, but would cause problems for people.

Paul Craig: What to do when you get pwned?

Kirk Jackson — Thu, 15 Jul 2010 00:00:10 GMT

Paul Craig works at security-assessment.com as a forensic investigator.

Forensic investigation: Fact-based investigation - must be reproducible and not based on anything subjective.

If you're going to get hacked, it will start at your web app. Firewalls generally stop all other traffic.

Treat all results as possible legal evidence - could be used for murder etc cases. Evidence could be used to allow police to arrest a suspect.

Most computer crimes in NZ will be tried under property law with a judge and jury.

All evidence may need to be provided to defendant to cast doubt on the evidence. How was it collected or analysed?

Common things customers say:

- Assumptions
- They only compromised one server - assume it has happened more than once
- We already dealt with it - probably destroyed all forensic evidence (could come back to bite in the future)
- It's too hard / not my problem

What to do when there's an incident:

How you act makes all the difference. Smooth engagements and do things as fast as possible.

Need a single point of contact for all security incidents within an organisation.

Appoint an incident response team - includng someone with internal clout, legal support.

Find a forensics supplier in advance. Don't leave it till when there's an incident.

It's a specialised industry, and you shouldn't do it yourself.

Media:

Media love a hacking story. This makes things stressful.

You need a bottom draw letter pre-written that you can give to the media. Get it signed by the CEO now.

Technical incident response:

Treat with urgency, gather incident team together in a secure location.

Get incident responder into the system as soon as possible to get current connections, arp caches etc.

- Disable scheduled patches, updates, restarts
- Unplug from internet / firewall it
- Leave the server powered on
- Put a big sign "Do not touch"

Within a day or less if possible.

Police reports:

If you have evidence that a crime has been committed, or something could be committed (e.g. fraud), file an incident report with police. As much evidence as possible.

Will you catch them?

If NZ / AU - likely.

If UN / NATO, possible but involved IPTF task force.

Other country: very slim chance of catching them.

When don't you have to file a report:

No loss of finances, no increase in fraud risk, no chance of repurcussions / fines.

How to do forensics:

Paul then talked about how security-assessment.com do forensics testing. Take-away: it's hard, and in order to provide evidence in court you won't actually be able to do it yourself.

Examples:

Paul gave examples of when they'd be engaged with customers. Problems encountered:

- They knew they had been hacked, but hadn't told each other
- Meeting in insecure places
- Taking too long to figure out what to do
- Companies that don't know how to respond
- Assuming evidence has been destroyed already

Without senior executive support, nothing will happen. Forensic and technical response isn't a technical problem: it is an entire business problem.

Take-home:

Sooner or later, you'll get hacked. When it happens, take it seriously.

Prepare for that incident straight away. Figure out what you'd do?

Stay cool when it happens, follow the game plan.

Never assume anything!

Questions:

How do you deal with situations where the hacked website needs to be back up in 10 minutes? So you don't have time to do forensics?

- Bring up a DR server if you have a safe backup.
- If it's compromised, you have to take it off immediately if someone is on that server at that time

How do you deal with virtualisation? When you don't have physical access to a machine?

- Can get all active memory and disk onto a disk
- Can take the entire VM snapshot and rebuild into a real computer again

What about if it's a cloud provider?

- Probably have no access to get an image. Comes down to whether we can get that access.

Does a live image impact the integrity of the evidence?

- Hash the evidence as soon as it is taken, so we can prove the image is unaltered.

If hacker uses anonymity services like tor / proxies?

- Often there's one request where they connect back directly.
- Often there's still some fragments of evidence remaining.
- Might be able to find out what they did, but not necessarily who did it.
- "Your credit cards have not been touched"

Roberto Suggi Liverani - Defending Against Application Level DoS Attacks

Kirk Jackson — Wed, 14 Jul 2010 23:37:58 GMT

Roberto's talk covered application-level vulnerabilities, and gave some ideas on how to plan for them, how to react when they happen, and how to recover from them.

Most denial of service attacks have traditionally covered the layer 3 or 4 (i.e. the transport or network stack), but Roberto has seen attacks against applications and web service layers.

Can lead to increased use of resources like CPU, network

Root causes:

- bug
- application logic open to abuse
- session level attacks

Examples:

PHP: Can create an unbounded size object in code

Failure to release resource: DB exception doesn't close connection. Attacker can cause app to open up lots of DB connections and deny service.

Sesion related: storing lots of session objects that consume resources, so attacker can target this to exhaust server resources.

User input as a loop counter: If the user can control how many times an expensive operation is performed, it can cause the app to do lots of demanding work.

=> Put in some limits, don't allow the user to set in their code.

Regular expressions: Certain input may cause lots of passes through a regular expression, causing lots of CPU to be used.

Other web problems can amplify DOS effects (XSS, XSRF, SQL injection, large file input)

Recommendations:

- Input strict validation and filtering
- Handle exceptions and properly release resources
- Set limits for:
- Session related objects
- Token expiration
- Object allocation
- Loop counters
- User registration - captcha
- Concurrent session tokens per IP address

- Testing your web app
- Test Regex, database queries
- DoS and stress testing
- Security testing

XML attacks:

There are lots of attacks against XML or web services.

Recommendations: don't use customised XML parser, input validation, use an XML firewall, limit the sizes of input messages, disable external DTDs.

Webserver attacks:

Attacks to use up all the threads on a webserver, or slow down the processing so the server can't process other requests.

Recommendations: Apache and IS have modules or configuration settings. Make sure you test the changes.

Database attacks:

Make the DB do more work than they should. E.g. cause a slow scan over a whole table, or avoid caching layers.

Recommendations: Input validation, captcha or user limits, only let authenticated users perform slow queries, use caching layers.

If you are under attack:

Be prepared, have a plan, simulate it often.

When under attack:

Is it real? What is the target? Is the target critical?

Reacting:

Several methods: slow down the attack, deflect it, drop connections, escalate to authorities or other nefarious ways to stop botnets.

Recovering:

Meet up to debrief as soon as possible afterwards. What lessons were learnt? Update incident plan.

What was the root cause? What if it happens again? Provide all data to law enforcement.

Conclusion:

No generic solution to DOS.

If offered a DOS solution product, look carefully before committing.

Start networking with people that can help you.

Brett Moore: Don't try this at home

Kirk Jackson — Wed, 14 Jul 2010 21:59:45 GMT

Brett presented a talk on some of the "Not so common code vulnerabilities".

The theme of his talk was that we shouldn't trust user input.

My notes:

A security vulnerability in an app - a weakness that allows a user to perform an action that was unintended.

AppTrends graph (cenzic.com) - input validation is the cause of everything (XSS, SQL injection, etc)

Frameworks won't protect you (e.g. .NET, PHP, Java frameworks).

Frameworks can promote bad practices, or have bugs in them themselves.

- Spring Framework http://blog.o0o.nu/ - override class loaded
- Struts2 - execute arbitrary java code

Examples of problems:

Trusting filenames / urls from the user

Using 302 Redirects as a security measure - returning secure

content below the redirect by mistake

Captchas: Tell whether it's a human or computer. Bad implementations where people have rolled their own and make it easy for computer to answer

Online shopping: Response from DPS comes in a browser redirect, so you can intercept it, and add extra stuff to the shopping cart after paying, but before the website thinks the order is finished.

Flash: Parameters for a flash movie can be entered in the url as well. Movie hosted on our site can end up displaying images or other content from our attack website.

Forgotten password: Stored proc truncates email address to 100 characters when looking up the user, but application uses the whole string. This can lead to an attacker receiving the forgotten password email.

Java object serialisation: Object is serialised into a cookie using Base64 encoding. Ooops: It contains something sensitive like a password.

PHP app in a security appliance used by a .mil: Shell out to a system command using a url parameter passed via an unauthenticated user.

Cookies: storing security data in a cookie - example of LoginAttempts - an attacker can modify the cookie to their hearts content.

Cookie: remember me functionality - store random token in the database and send it to the user as a cookie, so they can log in automatically. Vulnerability: flawed if null was stored in both the db and the cookie.

Lesson:

Never trust the users input

Input validation is the key.

You can use hidden form fields or cookies, as long as the backend input validation is secure. You can't trust that the frontend is doing things correctly.

Backend should:
- Validate the data
- Ensure the user is authorised to access the data

Data comes in many forms (upper / lower case, encoded etc)

- Decode the data, or reject it if a normal user wouldn't send it

Ensure data conforms to the correct format
- Check length, type, min / max values
- Alphanumeric / valid date only

Reject invalid data, rather than attempting to fix it up.

Beware writing your own data sanitisation functions - needs to be well tested and document. Use OWASP or language features if possible.

- Easy to write bad sanitisation. Examples of bad url testing,

XSS works without script

Takeaways:

- Review your code. Have "Code Review Parties"
- Have peer reviews
- Have standards, and stick to them

Questions to Brett:

Should we still trust CAPTCHA?

Still effective at the moment, but can be broken.

OWASP NZ: PCI-DSS for OWASP Practitioners: Dean Carter, security-assessment.com

Kirk Jackson — Mon, 13 Jul 2009 03:46:55 GMT

If you store, transmit or process credit card data, PCI applies.

How can OWASP help you with PCI compliance?

Credit card data:

Primary Account Number (PAN): Can store it, but protection required.
Can never store the CVD 3 digit number or mag stripe

Card data attacks have been increasing in sophistication.

PCI-DSS affects anyone who transmits, processes or stores payment card data. E.g. merchants, service providers (e.g. Paymark, DPS).

Look at 12 requirements of PCI-DSS (firewalls, storage etc)

Protecting stored data:

You must not store sensitive authentication data. Principle: if you don't need it, don't store it. Consider outsourcing, truncation, tokenisation.

Tokenisation: Replace PAN with a unique identifier "token"

Truncation: don't store all the data (e.g. first 4, last 4 digits)

Encryption: Encrypt at point of capture, only decrypt when required, use industry standard encryption, protect your keys.

Developing secure applications / Test app was built securely / Use secure coding guidelines:

Standard OWASP guidelines

Annual risk assessment:

Every year, new threats will affect your site. Go and re-assess against the new threats.

Fixing legacy systems: make sure no old data is lying around.

Real life example: it's very easy to mess up (example of reverting to old code)

Parting thoughts: achieve, maintain and validate compliance. Secure development is a key activity. OWASP is a good source. Reduce storage of PAN data.

OWASP NZ: Application Bug Chaining: Mark Piper, Catalyst IT

Kirk Jackson — Mon, 13 Jul 2009 02:57:28 GMT

Bug chaining - an idea that hasn't really propagated yet.

How do we rate how severe a bug is? Consider how easy it is to exploit, where it is accessible from (client-side, server-side, internet, local, mass exploitable, targeted exploit, etc).

Audience attempted to rate the severity of a couple of bugs:

SQL injection on authenticated site -> medium/high
File upload php files on authenticated site -> high/critical
Local file disclosure -> medium/high
XSS - reflective, authenticated -> low/medium

Is attacker considered 'authenticated' once there is an XSS attack? Any subsequent attacks can be treated as authenticated.

When you join together the XSS bug with the file upload bug, then it's critical!

Bug chaining: taking multiple bugs and chaining them together to create exploitable vulnerabilities. Instead of looking at each individual bug, look at how they can be combined together.

There are now frameworks to help chain together exploits - and this is how a lot of worms now work.

Recent examples of chaining exploits: PHPMyAdmin <= 3.1.3; SugarCRM <= 5.2.0e - compromise server through 3 bugs together.

How to deal with this? CVSSv2:

Common Vulnerability Scoring System v2.0
Scoring system for assessing bugs
Considers exploit complexity, application location, authentication, target likelihood etc
Can be very complex, time consuming, difficult to follow

"You can explain this stuff all day, but when network admins actually see you do it, that's when they understand" Brett Moore

VtigerCRM - large open-source CRM system which fixed problems with a security patch, but don't link to the fix (and haven't installed it themselves!).

He wrote a BeEf module for VtigerCRM that can run as an auto-run module (took less than 2 hours to write):

Chains file upload and XSS bug to upload a malicious PHP script to start a command shell
Connection is from server to the attackers machine, so user doesn't need to stay connected

Summary:

Don't look at severity of individual bugs - need to look at how bugs can be joined together.

Understand the bugs.

Follow the OWASP coding and testing guidelines.

Tools:

BeEf - command console for an attacker to run script on the client computer. Modular list of exploits, and control multiple victims. Autorun modules to automatically execute modules within 1.5-2 seconds.

OWASP NZ: Exploiting Firefox Extensions: Roberto Suggi Liverani & Nick Freeman, Security-Assessment.com

Kirk Jackson — Mon, 13 Jul 2009 02:19:53 GMT

Firefox extensions: They're just software, like ActiveX. Extend, modify and control the browser.

Firefox extension points:

XUL: XML user interface language
XBL: XML Binding Language - logical behaviour of widgets
XPCOM: Reusable components, interface to file system etc.
XPConnect: Allows Javascript to connect to XPCOM
Chrome: Special browser zone that is fully trusted by firefox - code is fully trusted, has access to filesystem, user passwords etc.

Mozilla security extension model is non-existent. All extensions are fully trusted by Firefox - no boundaries between extensions, they can modify each other without the user knowing. Can be coded in C++ and subject to memory corruption etc.

Extensions are very popular (billion downloads) and can be found everywhere - social networks, search engines, software packages (skype, anti-virus), anti-phishing toolbars.

Biggest problem is the human side of things - Addins.mozilla.org recommend extensions and add a 'recommended' icon next to them. Extension source code isn't read by third parties ("It's not the linux kernel").

There's no protection from an extension with a security problem, it will bypass any other phishing / malware protection extensions.

Extensions aren't signed (even the Mozilla ones), so we can't rely on people checking signatures.

If an extension is originally trusted, then subsequent updates won't go through the same review process.

No current guidelines for testing a Firefox extension, so security-assessement.com havce come up with their own methodology (whitepaper to be released this year, early next year):

Isolated testing: Only test one extension at a time, on different OSes with different Firefox versions.
Information gathering: How does the extension work, how is it installed? Look inside the extension package (a zip file) and look for malicious files (e.g. .exe, .msi etc)
Look for XPInstall API functions that are dangerous (e.g. executing code on install)
Look for suspicious files in the extension folder (e.g. softlinks to other directories)
Look inside install.rdf - some tags can hide extensions so they don't appear in the addon manager
Extensions can have the same description as other installed extensions, so two appear in addon manager
Does the extension try to trick the user into thinking it's verified?
Look for pointers outside the extension, or flags that expose the extension object or content to untrusted code (e.g. contentaccessible=yes or xpcnativewrappers=no)
Extensions can be merged into the firefox UI - e.g. top toolbar, bottom status bar. They can also modify existing buttons e.g. Reload, Back, Forward or Home button.
Use the extension. Check the DOM of a test page with the extension loaded (they used mozreply to do this)
Debugging: can set breakpoints using Javascript debugger.
Sandbox: can be sidestepped by replacing code inside the sandbox or evaluating it from outside
XPCOM components: .dll or .so - compiled code that the extension may ship with, or may use existing components on the machine. May need to review source code or decompile. A bunch of components to watch out for.
wrappedJSObject: removes the protection of the XPComComponent, so they are avoiding the firefox protection.
Watch out for callback functions, which may be replaced / modified
window.OpenDialog: Opens any URI with elevated chrome privileges
Auth: Some expose credentials in plain text, e.g. GET or basic auth
Auth: Some expose functionality via javascript that can side-step normal process
Skype extension - a javascript call that any web page can use to start dialing your skype to any
XSS: Watch out for XSS issues - can execute in the chrome zone from DOM events, embedded XSS, recursive iframes
XSS: Extensions loading external scripts

They have applied their methodology to different extensions, and some responses have been slow or non-existent!

Here are some extensions that were demoed and had problems. They are all common or Mozilla recommended (all these have been fixed):

FireFTP: Could include malicious code in the welcome method of an FTP server, and the browser would execute it. Showed a proof of concept sending the contents of win.ini to a different server, and using BeEf to control client.
CoolPreviews: Susceptible to XSS if a data:// URI is used. Showed a remote code execution when right-clicking on a link and previewing it with CoolPreviews.
WizzRSS: HTML and Javascript in the <description> tag of RSS feeds is executed in the chrome zone. Showed a reverse shell onto the Windows machine from a malicious users machine.

Extension developers and vendors haven't got a security disclosure process yet - they don't know how to deal with the issues yet. Some extensions don't even publish an email address for the author.

Tools:

Firebug
MozRepl
BeEf - command console for an attacker to run script on the client computer.

OWASP NZ: Testing Web Services: Nick van Dadelszen, Lateral Security

Kirk Jackson — Sun, 12 Jul 2009 23:47:27 GMT

With shift to web services, where we are relying on client to secure stuff, we have to remember not to trust the client.

Gave a methodology for testing web services:

Service discovery:

Look for WSDL or similar files that contain service info, using search engines, site spidering or looking at app behaviour

Method discovery:

Look inside the WSDL to see what methods are available, or if there isn't one, you can brute force the webservice with common method names to find ones that exist.

OWASP top 10. These still all apply to web service calls, including:

Malicious file execution, insecure direct object reference,
CSRF with AJAX clients
Information leakage
Broken auth and session mgmt
Insecure crypto storage
Insecure communications - SSL is important
Failure to restrict URL access - protect admin etc web services from anonymous access

Web service specific tests:

XML issues (external entities, malformed XML, recursive XML, XML entity expansion, XML attribute blowup, overlarge XML and CDATA injection)

Can find out details inside the secure network, and CSRF etc machines in there.

WS-Routing issues

WS-Security is not a panacea - secures the method integrity and confidentiality, but doesn't stop bad stuff coming through.

Tools shown:

SIFT web method search tool - brute force the web service to find out which methods are supported.
Foundstone WS Digger - automate attacks against web services (XSS, SQL, Xpath etc)
Webscarab - to modify XML posted to web services and try connecting to external entities

OWASP NZ: Vulnerabilities in Action: Brett Moore, Insomnia Security

Kirk Jackson — Sun, 12 Jul 2009 22:37:46 GMT

If you don't own the 3 OWASP books, you've failed.

We're still facing the same vulnerabilities we already have, because we are doing something wrong. Maybe it's security professionals that are doing something wrong, by not educating developers properly.

Big security companies still having problems with their websites.

Most vulnerabilities are well known.

Security people don't write code. developers do. They don't "get" security:

Don't fix the root cause
Don't understand the threat
Most have never seen a vulnerability exploited

Sitting down with developers and stepping them through a vulnerability helps show them the light and they understand and think about vulnerabilities.

Talk today designed to show developers exploits in action.

Tools showed:

Burp - proxy tool for intercepting requests
A custom sitemap tool that Insomnia uses
An MS-SQL Enumeration tool that takes a vulnerable url and pulls out all the DB info using the master db to enumerate tables
ASPX Spy - if you can get this ASP.NET file up on to a server and run, it provides a UI for playing around with the OS.
SQL Map - an automatic SQL injection tool - can enumerate the DB, even if the data is not displayed by inferring the state of the db based on the page output.

Problems shown:

Robots.txt is not a place to list parts of your site that you don't want people to know about :)
Buying -1 quantity of a $1000 book leads to the users credit on the shopping site increasing by $1000 :)
XML parsing vulnerability that allows external entities to be referenced in the XML provided to a web service - which can pull the contents of a file off the server.
Query string parameters passed to the command interpreter, and used for file names.
PHP include let's you include PHP source from another web server (looks like you need to disable URL fopen wrappers).
Only securing GET requests to an admin directory.
Showed a fake version of the CCIP website with multiple problems.
Admin interface for a website is exposed to the internet.

Open questions:

Who owns server configuration? Architects, developers, system administrators? If server or framework config changes, then we're insecure.
Is it security professionals job to make sure problems are corrected?

OWASP NZ: Insecurity and the Internet: Paul Craig – Security-Assessment.com

Kirk Jackson — Sun, 12 Jul 2009 21:44:40 GMT

Paul raised the question: "Is internet security getting better or worse?"

By 2004 we had bought lots of security products, and now only port 80 is the only open port (default DENY). Hackers started hacking web apps instead.

Classic ASP was easy to hack. until in 2005 when vendors started releasing safer technology frameworks (2005? We were using it in 2002)

Note: ASP.NET doesn't have XSS protection built in, unless you leave ValidateRequest on (which no-one does), as controls only sporadically escape their output.

Paul looked at Security-Assessment's old pen-test projects and compared their vulnerabilities to those run recently.

"In 2003-2005, web application developers were F$%^&* bad"

"Developers fail at anything to do with files"

But the situations hasn't got much better lately. Admin sections are still accessible, SQL injection still found, but less common, file uploads allowing directory traversal.

When developers use framework security controls, they're okay. If they use custom security code, they mess it up.

"Less vulnerabilities in 2009 resulted in a shell"

"Security only works flawlessly when it's already implemented in the framework" - when developers build their own code, they normally mess it up.

Summary: The internet is getting more secure, but we're not there yet! Only need one bug to get in to a system.