1. Why Is DDoS Hard to Solve? 1.A simple form of attack 2.Designed to prey on the Internet’s strengths 3.Easy availability of attack machines 4.Attack can look like normal traffic 5.Lack of Internet enforcement tools 6.Hard to get cooperation from others 7.Effective solutions hard to deploy

2. 1. Simplicity Of Attack Basically, just send someone a lot of traffic More complicated versions can add refinements, but that’s the crux of it No need to find new vulnerabilities No need to worry about timing, tracing, etc. Toolkits are readily available to allow the novice to perform DDoS Even distributed parts are very simple

3. 2. Preys On Internet’s Strengths The Internet was designed to deliver lots of traffic – From lots of places, to lots of places DDoS attackers want to deliver lots of traffic from lots of places to one place Any individual packet can look proper to the Internet Without sophisticated analysis, even the entire flow can appear proper

4. Internet Resource Utilization Internet was not designed to monitor resource utilization – Most of it follows first come, first served model Many network services work the same way And many key underlying mechanisms do, too Thus, if a villain can get to the important resources first, he can often deny them to good users

5. 3. Availability Of Attack Machines DDoS is feasible because attackers can enlist many machines Attackers can enlist many machines because many machines are readily vulnerable Not hard to find 1,000 crackable machines on the Internet – Particularly if you don’t care which 1,000 Botnets numbering hundreds of thousands of hosts have been discovered

6. Can’t We Fix These Vulnerabilities? DDoS attacks don’t really harm the attacking machines Many people don’t protect their machines even when the attacks can harm them Why will they start protecting their machines just to help others? Altruism has not yet proven to be a compelling argument for for network security

7. 4. Attacks Resemble Normal Traffic A DDoS attack can consist of vast number of requests for a web server’s home page No need for attacker to use particular packets or packet contents So neat filtering/signature tools may not help Attacker can be arbitrarily sophisticated at mirroring legitimate traffic – In principle – Not often done because dumb attacks work so well

8. 5. Lack Of Enforcement Tools DDoS attackers have never been caught by tracing or observing attack Only by old-fashioned detective work – Really, only when they’re dumb enough to boast about their success The Internet offers no help in tracing a single attack stream, much less multiple ones Even if you trace them, a clever attacker leaves no clues of his identity on those machines

9. What Is the Internet Lacking? No validation of IP source address No enforcement of amount of resources used No method of tracking attack flows – Or those controlling attack flows No method of assigning responsibility for bad packets or packet streams No mechanism or tools for determining who corrupted a machine

10. 6. Poor Cooperation In the Internet It’s hard to get anyone to help you stop or trace or prevent an attack Even your ISP might not be too cooperative Anyone upstream of your ISP is less likely to be cooperative – ISPs more likely to cooperate with each other, though Even if cooperation occurs, it occurs at human timescales – The attack might be over by the time you figure out who to call

11. 7. Effective Solutions Hard To Deploy The easiest place to deploy defensive systems is near your own machine – Defenses there might not work well (firewall example) There are effective solutions under research – But they require deployment near attackers or in the Internet core – Or, worse, in many places A working solution is useless without deployment – Hard to get anything deployed if deploying site gets no direct advantage

12. Resource Limitations Don’t allow an individual attack machine to use many of a target’s resources Requires: – Authentication, or – Making the sender do special work (puzzles) Authentication schemes are often expensive for the receiver Existing legitimate senders largely not set up to handle doing special work Can still be overcome with a large enough army of zombies

13. Hiding From the Attacker Make it hard for anyone but legitimate clients to deliver messages at all E.g., keep your machine’s identity obscure A possible solution for some potential targets – But not for others, like public web servers To the extent that approach relies on secrecy, it’s fragile – Some such approaches don’t require secrecy

14. Resource Multiplication As attacker demands more resources, supply them Essentially, never allow resources to be depleted Not always possible, usually expensive Not clear that defender can keep ahead of the attacker But still a good step against limited attacks More advanced versions might use Akamai-like techniques

15. Trace and Stop Attacks Figure out which machines attacks come from Go to those machines (or near them) and stop the attacks Tracing is trivial if IP source addresses aren’t spoofed – Tracing may be possible even if they are spoofed May not have ability/authority to do anything once you’ve found the attack machines Not too helpful if attacker has a vast supply of machines

16. Filtering Attack Streams The basis for most defensive approaches Addresses the core of the problem by limiting the amount of work presented to target Key question is: – What do you drop? Good solutions drop all (and only) attack traffic Less good solutions drop some (or all) of everything

17. Filtering Vs. Rate Limiting Filtering drops packets with particular characteristics – If you get the characteristics right, you do little collateral damage – At odds with the desire to drop all attack traffic Rate limiting drops packets on basis of amount of traffic – Can thus assure target is not overwhelmed – But may drop some good traffic You can combine them (drop traffic for which you are sure is suspicious, rate-limit the rest) but you gain a little

18. Where Do You Filter? Near the target? Near the source? In the network core? In multiple places?

19. Filtering Location Choices Near target Near source In core

20. Filtering Location Choices Near target – Easier to detect attack – Sees everything – May be hard to prevent collateral damage – May be hard to handle attack volume Near source In core

21. Filtering Location Choices Near target Near source – May be hard to detect attack – Doesn’t see everything – Easier to prevent collateral damage – Easier to handle attack volume In core

22. Filtering Location Choices Near target Near source In core – Easier to handle attack volume – Sees everything (with sufficient deployment) – May be hard to prevent collateral damage – May be hard to detect attack

23. How Do You Detect Attacks? Have database of attack signatures Detect anomalous behavior – By measuring some parameters for a long time and setting a baseline Detecting when their values are abnormally high – By defining which behavior must be obeyed starting from some protocol specification

24. How Do You Filter? Devise filters that encompass most of anomalous traffic Drop everything but give priority to legitimate- looking traffic – It has some parameter values – It has certain behavior

25. DDoS Defense Challenges Need for a distributed response Economic and social factors Lack of detailed attack information Lack of defense system benchmarks Difficulty of large-scale testing Moving target

26. TCP SYN Flood Attacker sends lots of TCP SYN packets – Victim sends an ack, allocates space in memory – Attacker never replies – Goal is to fill up memory before entries time out and get deleted Usually spoofed traffic – Otherwise patterns may be used for filtering – OS at the attacker or spoofed address may send RST and free up memory

27. TCP SYN Cookies Effective defense against TCP SYN flood – Victim encodes connection information and time in ACK number – Must be hard to craft values that get encoded into the same ACK number – use crypto for encoding – Memory is only reserved when final ACK comes Only the server must change – But TCP options are not supported – And lost SYN ACKs are not repeated

28. Small-Packet Floods Overwhelm routers – Create a lot of pps – Exhaust CPU – Most routers can’t handle full bandwidth’s load of small packets No real solution, must filter packets somehow to reduce router load

29. Shrew Attack Periodically slam the victim with short, high-volume pulses – Lead to congestion drops on client’s TCP traffic – TCP backs off – If loss is large back off to 1 MSS per RTT – Attacker slams again after a few RTTs Solution requires TCP protocol changes – Tough to implement since clients must be changed

30. Flash-Crowd Attack Generate legitimate application traffic to the victim – E.g., DNS requests, Web requests – Usually not spoofed – If enough bots are used no client appears too aggressive – Really hard to filter since both traffic and client behavior seem identical between attackers and legitimate users

31. Reflector Attack Generate service requests to public servers spoofing the victim’s IP – Servers reply back to the victim overwhelming it – Usually done for UDP and ICMP traffic (TCP SYN flood would only overwhelm CPU if huge number of packets is generated) – Often takes advantage of amplification effect – some service requests lead to huge replies; this lets attacker amplify his attack

32. Sample Research Defenses Pushback Traceback SOS Proof-of-work systems Human behavior modeling

33. Pushback 1 Goal: Preferentially drop attack traffic to relieve congestion Local ACC: Enable core routers to respond to congestion locally by: – Profiling traffic dropped by RED – Identifying high-bandwidth aggregates – Preferentially dropping aggregate traffic to enforce desired bandwidth limit Pushback: A router identifies the upstream neighbors that forward the aggregate traffic to it, requests that they deploy rate-limit 1 ”Controlling high bandwidth aggregates in the network,” Mahajan, Bellovin, Floyd, Paxson, Shenker, ACM CCR, July 2002

34. Can it Work? Even a few core routers are able to control high- volume attacks Separation of traffic aggregates improves current situation – Only traffic for the victim is dropped – Drops affect a portion containing the attack traffic Likely to successfully control the attack, relieving congestion in the Internet Will inflict collateral damage on legitimate traffic

35. Advantages and Limitations +Routers can handle high traffic volumes +Deployment at a few core routers can affect many traffic flows, due to core topology +Simple operation, no overhead for routers +Pushback minimizes collateral damage by placing response close to the sources –Pushback only works in contiguous deployment –Collateral damage is inflicted by response, whenever attack is not clearly separable –Requires modification of existing core routers 35

36. Traceback 1 Goal: locate the agent machines Each packet header may carry a mark, containing: – EdgeID (IP addresses of the routers) specifying an edge it has traversed – The distance from the edge Routers mark packets probabilistically If a router detects half-marked packet (containing only one IP address) it will complete the mark Victim under attack reconstructs the path from the marked packets 1 “Practical network support for IP Traceback,” Savage, Wetherall, Karlin, Anderson, ACM SIGCOMM 2000

37. Traceback and IP Spoofing Traceback does nothing to stop DDoS attacks It only identifies attackers’ true locations – Comes to a vicinity of attacker If IP spoofing were not possible in the Internet, traceback would not be necessary There are other approaches to filter out spoofed traffic

38. Can it Work? Incrementally deployable, a few disjoint routers can provide beneficial information Moderate router overhead (packet modification) A few thousand packets are needed even for long path reconstruction Does not work well for highly distributed attacks Path reassembly is computationally demanding, and is not 100% accurate: – Path information cannot be used for legal purposes – Routers close to the sources can efficiently block attack traffic, minimizing collateral damage

39. Advantages and Limitations +Incrementally deployable +Effective for non-distributed attacks and for highly overlapping attack paths +Facilitates locating routers close to the sources –Packet marking incurs overhead at routers, must be performed at slow path –Path reassembly is complex and prone to errors –Reassembly of distributed attack paths is prohibitively expensive

40. SOS 1 Goal: route only “verified user” traffic to the server, drop everything else Clients use overlay network to reach the server Clients are authenticated at the overlay entrance, their packets are routed to proxies Small set of proxies are “approved” to reach the server, all other traffic is heavily filtered out 40 1 “ SOS: Secure Overlay Services, ” Keromytis, Misra, Rubensteain, ACM SIGCOMM 2002

41. SOS User first contacts nodes that can check its legitimacy and let him access the overlay – access points An overlay node uses Chord overlay routing protocol to send user’s packets to a beacon Beacon sends packets to a secret servlet Secret servlets tunnel packets to the firewall Firewall only lets through packets with an IP of a secret servlet – Secret servlet’s identity has to be hidden, because their source address is a passport for the realm beyond the firewall – Beacons are nodes that know the identity of secret servlets If a node fails, other nodes can take its role 41

42. Can It Work? SOS successfully protects communication with a private server: – Access points can distinguish legitimate from attack communications – Overlay protects traffic flow – Firewall drops attack packets Redundancy in the overlay and secrecy of the path to the target provide security against DoS attacks on SOS 42

43. Advantages And Limitations +Ensures communication of “verified user” with the victim +Resilient to overlay node failure +Resilient to DoS on the defense system –Does not work for public service –Traffic routed through the overlay travels on suboptimal path –Brute force attack on links leading to the firewall still possible 43

44. Client Puzzles 1 Goal: defend against connection depletion attacks When under attack: – Server distributes small cryptographic puzzles to clients requesting service – Clients spend resources to solve the puzzles – Correct solution, submitted on time, leads to state allocation and connection establishment – Non-validated connection packets are dropped Puzzle generation is stateless Client cannot reuse puzzle solutions Attacker cannot make use of intercepted packets 44 1 “Client puzzles: A cryptographic countermeasure against connection depletion attacks, ” Juels, Brainard, NDSS 1999

45. Can It Work? Client puzzles guarantee that each client has spent a certain amount of resources Server determines the difficulty of the puzzle according to its resource consumption – Effectively server controls its resource consumption Protocol is safe against replay or interception attacks Other flooding attacks will still work 45

46. Advantages And Limitations +Forces the attacker to spend resources, protects server resources from depletion +Attacker can only generate a certain number of successful connections from one agent machine +Low overhead on server –Requires client modification –Will not work against highly distributed attacks –Will not work against bandwidth consumption attacks (Defense By Offense paper changes this) 46

47. Human Behavior Modeling 1 Goal: defend against flash-crowd attacks on Web servers Model human behavior along three dimensions – Dynamics of interaction with server (trained) Detect aggressive clients as attackers – Semantics of interaction with server (trained) Detect clients that browse unpopular content or use unpopular paths as attackers – Processing of visual and textual cues Detect clients that click on invisible or uninteresting links as attackers 47 1 “Modeling Human Behavior for Defense Against Flash Crowd Attacks”, Oikonomou, Mirkovic 2009.

48. Can It Work? Attackers can bypass detection if they – Act non-aggressively – Use each bot for just a few requests, then replace it But this forces attacker to use many bots – Tens to hundreds of thousands – Beyond reach of most attackers Other flooding attacks will still work 48

49. Advantages And Limitations +Transparent to users +Low false positives and false negatives –Requires server modification –Server must store data about each client –Will not work against other flooding attacks –May not protect services where humans do not generate traffic, e.g., DNS 49

50. Worms 50

51. Viruses don’t break into your computer – they are invited by you – They cannot spread unless you run infected application or click on infected attachment – Early viruses spread onto different applications on your computer – Contemporary viruses spread as attachments through E-mail, they will mail themselves to people from your addressbook Worms break into your computer using some vulnerability, install malicious code and move on to other machines – You don’t have to do anything to make them spread 51 Viruses vs. Worms

52. A program that: – Scans network for vulnerable machines – Breaks into machines by exploiting the vulnerability – Installs some piece of malicious code – backdoor, DDoS tool – Moves on Unlike viruses – Worms don’t need any user action to spread – they spread silently and on their own – Worms don’t attach themselves onto other programs – they exist as a separate code in memory Sometimes you may not even know your machine has been infected by a worm 52 What is a Worm?

53. They spread extremely fast They are silent Once they are out, they cannot be recalled They usually install malicious code They clog the network 53 Why Are Worms Dangerous?

54. Robert Morris, a PhD student at Cornell, was interested in network security He created the first worm with a goal to have a program live on the Internet in Nov. 1988 – Worm was supposed only to spread, fairly slowly – It was supposed to take just a little bit of resources so not to draw attention to itself – But things went wrong … Worm was supposed to avoid duplicate copies by asking a computer whether it is infected – To avoid false “yes” answers, it was programmed to duplicate itself every 7 th time it received “yes” answer – This turned out to be too much 54 First Worm Ever – Morris Worm

55. It exploited four vulnerabilities to break in – A bug in sendmail – A bug in finger deamon – A trusted hosts feature (/etc/.rhosts) – Password guessing Worm was replicating at a much faster rate than anticipated At that time Internet was small and homogeneous (SUN and VAX workstations running BSD UNIX) It infected around 6,000 computers, one tenth of then-Internet, in a day 55 First Worm Ever – Morris Worm

56. People quickly devised patches and distributed them (Internet was small then) A week later all systems were patched and worm code was removed from most of them No lasting damage was caused Robert Morris paid $10,000 fine, was placed on probation and did some community work Worm exposed not only vulnerabilities in UNIX but moreover in Internet organization Users didn’t know who to contact and report infection or where to look for patches 56 First Worm Ever – Morris Worm

57. In response to Morris Worm DARPA formed CERT (Computer Emergency Response Team) in November 1988 – Users report incidents and get help in handling them from CERT – CERT publishes security advisory notes informing users of new vulnerabilities that need to be patched and how to patch them – CERT facilitates security discussions and advocates better system management practices 57 First Worm Ever – Morris Worm

58. Spread on July 12 and 19, 2001 Exploited a vulnerability in Microsoft Internet Information Server that allows attacker to get full access to the machine (turned on by default) Two variants – both probed random machines, one with static seed for RNG, another with random seed for RNG (CRv2) CRv2 infected more than 359,000 computers in less than 14 hours – It doubled in size every 37 minutes – At the peak of infection more than 2,000 hosts were infected each minute 58 Code Red

59. 59 Code Red v2

60. 43% of infected machines were in US 47% of infected machines were home computers Worm was programmed to stop spreading at midnight, then attack www1.whitehouse.gov – It had hardcoded IP address so White House was able to thwart the attack by simply changing the IP address-to-name mapping Estimated damage ~2.6 billion 60 Code Red v2

61. Spread on January 25, 2003 The fastest computer worm in history – It doubled in size every 8.5 seconds. – It infected more than 90% of vulnerable hosts within 10 minutes – It infected 75,000 hosts overall Exploited buffer overflow vulnerability in Microsoft SQL server, discovered 6 months earlier 61 Sapphire/Slammer Worm

62. No malicious payload The aggressive spread had severe consequences – Created DoS effect – It disrupted backbone operation – Airline flights were canceled – Some ATM machines failed 62 Sapphire/Slammer Worm

63. 63 Sapphire/Slammer Worm

64. Both Slammer and Code Red 2 use random scanning o Code Red uses multiple threads that invoke TCP connection establishment through 3-way handshake – must wait for the other party to reply or for TCP timeout to expire o Slammer packs its code in single UDP packet – speed is limited by how many UDP packets can a machine send o Could we do the same trick with Code Red? Slammer authors tried to use linear congruential generators to generate random addresses for scanning, but programmed it wrong 64 Why Was Slammer So Fast?

65. 43% of infected machines were in US 59% of infected machines were home computers Response was fast – after an hour sites started filtering packets for SQL server port 65 Sapphire/Slammer Worm

66. 66 BGP Impact of Slammer Worm

67. 67 Stuxnet Worm Discovered in June/July 2010 Targets industrial equipment Uses Windows vulnerabilities (known and new) to break in Installs PLC (Programmable Logic Controller) rootkit and reprograms PLC – Without physical schematic it is impossible to tell what’s the ultimate effect Spread via USB drives Updates itself either by reporting to server or by exchanging code with new copy of the worm

68. Many worms use random scanning This works well only if machines have very good RNGs with different seeds Getting large initial population represents a problem – Then the infection rate skyrockets – The infection eventually reaches saturation since all machines are probing same addresses 68 Scanning Strategies “Warhol Worms: The Potential for Very Fast Internet Plagues”, Nicholas C Weaver

69. 69 Random Scanning

70. Worm can get large initial population with hitlist scanning Assemble a list of potentially vulnerable machines prior to releasing the worm – a hitlist – E.g., through a slow scan When the scan finds a vulnerable machine, hitlist is divided in half and one half is communicated to this machine upon infection – This guarantees very fast spread – under one minute! 70 Scanning Strategies

71. 71 Hitlist Scanning

72. Worm can get prevent die-out in the end with permutation scanning All machines share a common pseudorandom permutation of IP address space Machines that are infected continue scanning just after their point in the permutation – If they encounter already infected machine they will continue from a random point Partitioned permutation is the combination of permutation and hitlist scanning – In the beginning permutation space is halved, later scanning is simple permutation scan 72 Scanning Strategies

73. 73 Permutation Scanning

74. Worm can get behind the firewall, or notice the die-out and then switch to subnet scanning Goes sequentially through subnet address space, trying every address 74 Scanning Strategies

75. Several ways to download malicious code – From a central server – From the machine that performed infection – Send it along with the exploit in a single packet 75 Infection Strategies

76. Three factors define worm spread: – Size of vulnerable population Prevention – patch vulnerabilities, increase heterogeneity – Rate of infection (scanning and propagation strategy) Deploy firewalls Distribute worm signatures – Length of infectious period Patch vulnerabilities after the outbreak Worm Defense

77. This depends on several factors: – Reaction time – Containment strategy – address blacklisting and content filtering – Deployment scenario – where is response deployed Evaluate effect of containment 24 hours after the onset How Well Can Containment Do? “Internet Quarantine: Requirements for Containing Self-Propagating Code”, Proceedings of INFOCOM 2003, D. Moore, C. Shannon, G. Voelker, S. Savage

78. How Well Can Containment Do? Code Red Idealized deployment: everyone deploys defenses after given period

79. How Well Can Containment Do? Depending on Worm Aggressiveness Idealized deployment: everyone deploys defenses after given period

80. How Well Can Containment Do? Depending on Deployment Pattern

81. Reaction time needs to be within minutes, if not seconds We need to use content filtering We need to have extensive deployment on key points in the Internet How Well Can Containment Do?

82. Monitor outgoing connection attempts to new hosts When rate exceeds 5 per second, put the remaining requests in a queue When number of requests in a queue exceeds 100 stop all communication Detecting and Stopping Worm Spread “Implementing and testing a virus throttle”, Proceedings of Usenix Security Symposium 2003, J. Twycross, M. Williamson

83. Detecting and Stopping Worm Spread

85. Organizations share alerts and worm signatures with their “friends” – Severity of alerts is increased as more infection attempts are detected – Each host has a severity threshold after which it deploys response Alerts spread just like worm does – Must be faster to overtake worm spread – After some time of no new infection detections, alerts will be removed Cooperative Strategies for Worm Defense “Cooperative Response Strategies for Large-Scale Attack Mitigation”, Proceedings of DISCEX 2003, D. Norjiri, J. Rowe, K. Levitt

86. As number of friends increases, response is faster Propagating false alarms is a problem Cooperative Strategies for Worm Defense

87. Early detection would give time to react until the infection has spread The goal of this paper is to devise techniques that detect new worms as they just start spreading Monitoring: – Monitor and collect worm scan traffic – Observation data is very noisy so we have to filter new scans from Old worms’ scans Port scans by hacking toolkits Early Worm Detection C. C. Zou, W. Gong, D. Towsley, and L. Gao. "The Monitoring and Early Detection of Internet Worms," IEEE/ACM Transactions on Networking.

88. Detection: – Traditional anomaly detection: threshold-based Check traffic burst (short-term or long-term). Difficulties: False alarm rate – “Trend Detection” Measure number of infected hosts and use it to detect worm exponential growth trend at the beginning Early Worm Detection

89. Worms uniformly scan the Internet – No hitlists but subnet scanning is allowed Address space scanned is IPv4 Assumptions

90. Simple epidemic model: Worm Propagation Model Detect worm here. Should have exp. spread

91. Monitoring System

92. Provides comprehensive observation data on a worm’s activities for the early detection of the worm Consists of : – Malware Warning Center (MWC) – Distributed monitors Ingress scan monitors – monitor incoming traffic going to unused addresses Egress scan monitors – monitor outgoing traffic Monitoring System

93. Ingress monitors collect: – Number of scans received in an interval – IP addresses of infected hosts that have sent scans to the monitors Egress monitors collect: – Average worm scan rate Malware Warning Center (MWC) monitors: – Worm’s average scan rate – Total number of scans monitored – Number of infected hosts observed Monitoring System

94. MWC collects and aggregates reports from distributed monitors If total number of scans is over a threshold for several consecutive intervals, MWC activates the Kalman filter and begins to test the hypothesis that the number of infected hosts follows exponential distribution Worm Detection

95. Population: N=360,000, Infection rate:  = 1.8/hour, Scan rate  = 358/min, Initially infected: I 0 =10 Monitored IP space 2 20, Monitoring interval:  = 1 minute Code Red Simulation Infected hosts  estimation

96. Population: N=100,000 Scan rate  = 4000/sec, Initially infected: I 0 =10 Monitored IP space 2 20, Monitoring interval:  = 1 second Slammer Simulation Infected hosts  estimation