1. How to Own the Internet in Your Spare Time (Stuart Staniford Vern Paxson Nicholas Weaver ) Giannis Kapantaidakis University of Crete CS558

2. What could you do if you 0wn’d a million hosts? ► Distributed DOS attacks ► Access sensitive information ► Confuse-Corrupt the information Makes it valuable tool in Cyber warfare

3. How to 0wn a million hosts? Worms ► Programs that self-propagate across the Internet exploiting security flaws in widely- used services (As opposed to viruses, which require user action to spread.) (As opposed to viruses, which require user action to spread.)

4. Code Red I ► Initial version released July 13, 2001. ► Exploited known bug in Microsoft IIS Web servers. ► But: failure to seed random number generator.All worms attempted to compromise the same sequence of hosts. ► Linear spread, didn’t get very far

5. Code Red I v2 ► Released July 19, 2001. ► Same codebase but:  random number generator correctly seeded.  DDoS payload targeting IP address of www.whitehouse.gov www.whitehouse.gov ► That night, Code Red dies (except for hosts with inaccurate clocks!) ► It just takes one of these to restart the worm come the first of the next month!

6. Random Constant Spread Model ► N: Total number of Vulnerable servers in Internet ► K: Initial Compromise Rate: Rate at which a infected host is able to infect new hosts at the start of the incident ► a: Proportion of machines already compromised ► T: Time at which the incident happens ► Equation: Nda = (Na)K(1-a)dt ► Solution: a = e (K(t-T)) / 1 + e (K(t-T)) ► Good enough model (Works for Code Red I)

7. ► K=1.8 T=11.9 ► Max probe rate:510.000 scans per hour ► Came close to saturation before turning off

8. ► Reawake on Aug 1 st, K=0.7 ► Number of vulnerable systems was less than 40% as many as the first time ► Code Red more or less followed the model.

9. ► Released August 4, 2001. ► Comment in code: “Code Red II.”But in fact completely different code base. ► Payload: a root backdoor allowing unrestricted remote access ► Bug: crashes NT, only works right on Windows 2000. ► Used localized scanning strategy Code Red II

10. ► Attempt to infect addresses close to it  With probability 3/8 it chooses a random IP from with the class B address space of the infected machine  With probability ½ from class A  And with probability 1/8 from the whole internet ► Localized spreading works - hosts around it are often similar,topologically faster,spreads fast in internal network once it gets through the firewall Localized Scanning

11. Nimda ► Released September 18, 2001. ► Multi- mode spreading:  attack IIS servers via infected clients.  email itself to address book as a virus  copy itself across open network shares  modifying Web pages on infected servers in order to infect clients  scanning for Code Red II and sadmind backdoors (!)

12. ► Average - 100 connections per second ► About 3X number of Code Red probes ► Full functionality still not known!

13. ► Since Nimda spreads by multiple vectors,the counts shown for it may be an underestimate

14. ► Why Red Code I continues to gain strength each month remains unknown

15. Ways of reducing time ► Hit List scanning ► Permutation scanning ► Topological Scanning ► Internet scale hit-lists

16. Hit List scanning Idea: reduce slow startup phase. Idea: reduce slow startup phase. ► The author of the worm collects the list of around 10,000 - 50,000 potentially vulnerable machines ideally the ones with very good network connection, before releasing the worm ► The worm when released initially attacks these machines.So the initial infection is higher.When it infects a machine it divides the hit-list in half

17. Ways to get Hit list  Distributed Scanning - use zombies  Stealthy Scan- spread it over several months  DNS searches - e. g., www. domain. com  Spiders - ask the search engines  Just Listening-P2P, or exploit existing worms

18. Permutation Scanning Idea: reduce redundant scanning. ► Permutation allows a worm to detect when a host is already infected. ► Worms share a common permutation of the IP address space. ► An infected machine starts scanning just after their position in the permutation. When the worm sees an infected machine is chooses a new random start point.

19. Warhol Worm ► Based on:  Hit List &  Permutation Scanning ► Simulation Environment ► Results of Simulation

20. ► So now we already have methods to in <15 minutes. ► So now we already have methods to attack most vulnerable targets in <15 minutes.

21. Topological Scanning ► Alternative to hit-list scanning ► Use addresses available on victim’s machines. ► Use this as a start point before using Permutation Scanning. ► Peer to peer systems are highly vulnerable to this kind of scanning

22. ► Idea: use an Internet- sized hit list. ( entire address space scan roughly 2hr) ( entire address space scan roughly 2hr) ► Initial copy of the worm has the entire hit list. ► Each generation, infects n from the list, gives each 1/n. (Or, point them to a well- connected servers that serves up portions of the list.) ► If n=10 requires 7 generations to infect 10^7 hosts (less than 30 seconds! ) Flash Worms:The Real Danger

23. ► All those worms use singular communication patterns ► This forms the basis for automatic detection ► How can we remove that weakness from worms? Still need better worms

24. Contagion Worms ► Suppose you have two exploits:  Es : exploit in web server  Ec: exploit in client ► You infect a server (or client) with Es (Ec) ► Then you…wait. (Perhaps you bait, e. g., host porn.) ► When vulnerable client arrives, infect it. ► You send over both Es and Ec ► As client happens to visit other vulnerable servers infects ► Clearly there are no unusual communication patterns to be observed (other than slightly larger- than- usual transfers)

25. ► They become Dangerous with P2P systems because:  Likely only need a single exploit, not a pair.  Often, peers running identical software.  Often used to transfer large files.  Often give access to user’s desktop rather than server.  and can be Very Large Contagion Worms

26. ► KazaA: 9 million distinct IP connections with university hosts (5800) in a single month ► If you 0wn’d a single university, then in November, 2001 you could have 0wn’d 9 million additional hosts. ► How fast? Faster than 1 month. Contagion Worms

27. Updating and control ► Distributed control  Each worm has a list of other copies  Ability to create encrypted communication channels to spread info  Commands cryptographically signed by author.  Each worm copy, confirms signature,spreads to other copies and then executes the command ► Programmatic Updates  Operating systems allow dynamic code loading  New encrypted attack modules from Worm author

28. Centre for Disease Control ► Roles it is expected to perform  Identifying outbreaks  Rapidly Analyzing pathogens  Fighting Infections  Anticipating new vectors  Resisting future threats

29. How open? ► Have a open website (accessible to all)? ► Drawbacks:  Attacker targets the site  How correct an information placed on site is  Attacker also gains understanding  Some sources may not be willing to make their information public ► How International.