CAP 6135 Malware and Software Vulnerability Analysis Worms

CAP 6135: Malware and Software Vulnerability Analysis Worms Cliff Zou Spring 2012

Acknowledgement q This lecture uses some contents from the lecture notes from: q Dr. Vitaly Shmatikov CS 378 - Network Security and Privacy 2

Viruses vs. Worms VIRUS q q Propagates by infecting other programs Usually inserted into host code (not a standalone program) WORM q q Propagates automatically by copying itself to target systems Is a standalone program Sometime it is hard to distinguish virus or worm 3

Morris Worm q 1988: No malicious payload, but bogged down infected machines by uncontrolled spawning q q Infected 10% of all Internet hosts at the time Multiple propagation vectors q Remote execution using rsh and cracked passwords q q Buffer overflow in fingerd on VAX q q Tried to crack passwords using small dictionary and publicly readable password file; targeted hosts from /etc/hosts. equiv Standard stack smashing exploit DEBUG command in Sendmail q Buffer overflow attack Dictionary attack In early Sendmail versions, possible to execute a command on a remote machine by sending an SMTP (mail transfer) message 4

Worm propagation process q Find new targets q n Compromise targets u u n IP random scanning Exploit vulnerability Trick users to run malicious code -Spam Newly infected join infection army 5

Worm research motivation q q q Code Red (Jul. 2001) : 360, 000 infected in 14 hours Slammer (Jan. 2003) : 75, 000 infected in 10 minutes Congested parts of Internet (ATMs down…) Blaster (Aug. 2003) : 150, 000 ~ 8 million infected DDOS attack (shut down domain windowsupdate. com) Witty (Mar. 2004) : 12, 000 infected in half an hour Attack vulnerability in ISS security products Sasser (May 2004) : 500, 000 infected within two days Storm (Jan 2007): infected 1 to 5 millions computers 6

How do worms propagate? q Scanning worms q q Coordinated scanning q q Ask server for hosts to infect (e. g. , Google for “powered by phpbb”) Topological worm: q q Assemble tree of vulnerable hosts in advance, propagate along tree Meta-server worm q q Different worm instances scan different addresses Flash worms q q Worm chooses “random” address Use information from infected hosts (web server logs, email address books, config files, SSH “known hosts”) Contagion worm q Propagate parasitically along with normally initiated communication 7

Summer of 2001 [from “How to 0 wn the Internet in Your Spare Time”] Three major worm outbreaks 8

Code Red I q q q July 13, 2001: First worm of the modern era Exploited buffer overflow in Microsoft’s Internet Information Server (IIS) 1 st through 20 th of each month: spread q Find new targets by random scan of IP address space q q q Spawn 99 threads to generate addresses and look for IIS Creator forgot to seed the random number generator, and every copy scanned the same set of addresses 21 st through the end of each month: attack q Deface websites with “HELLO! Welcome to http: //www. worm. com! Hacked by Chinese!” 9

Exception Handling In IIS [See Chien and Szor, “Blended Attacks…”] q Overflow in a rarely used URL decoding routine q q q A malformed URL is supplied to vulnerable routine… … another routine notices that stack has been smashed and raises an exception. Exception handler is invoked… … the pointer to exception handler is located on stack. It has been overwritten to point to a certain instruction inside the routine that noticed the overflow… … that instruction is CALL EBX. At that moment, EBX is pointing into the overwritten buffer… … the buffer contains the code that finds the worm’s main body on the heap and executes it! 10

Code Red I v 2 q July 19, 2001: Same codebase as Code Red I, but fixed the bug in random IP address generation q q Compromised all vulnerable IIS servers on the Internet Large vulnerable population meant fast worm spread q q q Payload: distributed packet flooding (denial of service) attack on www. whitehouse. gov q q Scanned address space grew exponentially 350, 000 hosts infected in 14 hours!! Attack was on a fixed IP, so it was avoided. Still alive in the wild! 11

Code Red Code q q GET /default. ida? NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN %u 9090%u 6858%ucbd 3%u 7801%u 9090%u 8190%u 00 c 3 %u 0003%u 8 b 00%u 531 b%u 53 ff%u 0078%u 0000%u 00=a HTTP/1. 0 Detailed analysis at: http: //research. eeye. com/html/advisories/published/AL 20010717. ht ml 12

Simple worm propagation model q q q address space, size W N : total vulnerable It : infected by time t q q W N-It vulnerable at time t scan rate (per host), h Prob. of a scan hitting vulnerable # of increased infected in a unit time

Simple worm propagation

Propagation: Theory Classic epidemic model q q q N: total number of vulnerable hosts I(t): number of infected hosts at time t S(t): number of susceptible hosts at time t I(t) + S(t) = N b: infection rate Differential equation for I(t): d. I/dt = b. I(t) S(t) More accurate models adjust propagation rate over time Source: Cliff C. Zou, Weibo Gong, Don Towsley, and Lixin Gao. The Monitoring and Early Detection of Internet Worms, IEEE/ACM Transactions on Networking, 2005.

Code Red II q August 4, 2001: Same IIS vulnerability, completely different code, kills Code Red I q q q Scanning algorithm preferred nearby addresses q q q Known as “Code Red II” because of comment in code Worked only on Windows 2000, crashed NT Chose addresses from same class A with probability ½, same class B with probability 3/8, and randomly from the entire Internet with probability 1/8 Payload: installed root backdoor in IIS servers for unrestricted remote access Died by design on October 1, 2001 16

Nimda q September 18, 2001: Multi-modal worm using several propagation vectors q q q Exploit same IIS buffer overflow as Code Red I and II Bulk-email itself as an attachment to email addresses harvested from infected machines Copy itself across open network shares Add exploit code to Web pages on compromised sites to infect visiting browsers Scan for backdoors left by Code Red II Payload: turned-off code deleting all data on hard drives of infected machines 17

Signature-Based Defenses Don’t Help q q Nimda leaped firewalls! Many firewalls pass mail untouched, relying on mail servers to filter out infections q q q Most filters simply scan attachments for signatures (code snippets) of known viruses and worms Nimda was a brand-new infection with unknown signature, and scanners could not detect it Big challenge: detection of zero-day attacks q When a worm first appears in the wild, signature is not extracted until minutes or hours later 18

Code Red I and II (due to Vern Paxson) Code Red II dies off as programmed 19 With its predator gone, Code Red I comes back, still exhibiting monthly pattern

Slammer (Sapphire) Worm q January 24/25, 2003: UDP worm exploiting buffer overflow in Microsoft’s SQL Server q q Entire code fits into a single 404 -byte UDP packet q q Overflow was already known and patched by Microsoft… but not everybody installed the patch Worm binary followed by overflow pointer back to itself Classic buffer overflow combined with random scanning: once control is passed to worm code, it randomly generates IP addresses and attempts to send a copy of itself to port 1434 q MS-SQL listens at port 1434 20

Slammer Propagation q Scan rate of 55, 000 addresses per second q q q Initial infection was doubling in 8. 5 seconds (!!) q q Scan rate = rate at which worm generates IP addresses of potential targets Up to 30, 000 single-packet worm copies per second Doubling time of Code Red was 37 minutes Worm-generated packets saturated carrying capacity of the Internet in 10 minutes q q 75, 000 SQL servers compromised And that’s in spite of broken pseudo-random number generator used for IP address generation 21

05: 29: 00 UTC, January 25, 2003 [from Moore et al. “The Spread of the Sapphire/Slammer Worm”] 22

30 Minutes Later [from Moore et al. “The Spread of the Sapphire/Slammer Worm”] Size of circles is logarithmic in the number of infected machines 23

Slammer Impact q q $1. 25 Billion of damage Temporarily knocked out many elements of critical infrastructure q q q Bank of America ATM network Entire cell phone network in South Korea Five root DNS servers Continental Airlines’ ticket processing software The worm did not even have malicious payload… simply bandwidth exhaustion on the network and resource exhaustion on infected machines 24

Secret of Slammer’s Speed q Old-style worms (Code Red) spawn a new thread which tries to establish a TCP connection and, if successful, send a copy of itself over TCP q q Limited by latency of the network Majority of TCP connection requests will fail q q Slammer was a connectionless UDP worm q q q Each failed IP scan will take 21 seconds to finish (Windows, 3 tries) No connection establishment, simply send 404 -byte UDP packet to randomly generated IP addresses Limited only by bandwidth of the network A TCP worm can scan even faster q Dump zillions of 40 -byte TCP-SYN packets into link layer, send worm copy only if SYN-ACK comes back 25

Blaster and Welchia/Nachia q August 11, 2003: Scanning worm exploiting RPC service in Microsoft Windows XP and 2000 q First address at random, then sequential upward scan q q q Easy to detect, yet propagated widely and leaped firewalls Payload: denial of service against MS Windows Update + installing remotely accessible backdoor Welchia/Nachia was intended as a counter-worm q q Random-start sequential scan, use ICMP to determine if address is live, then copy itself over, patch RPC vulnerability, remove Blaster if found Did more damage by flooding networks with traffic 26

Search Worms q Generate search query q q q Search for version numbers of vulnerable software to find exploitable targets Search for popular domains to harvest email addresses Analyze search results q q [Provos et al. ] Remove duplicates, URLs belonging to search engine Infect identified targets q Reformat URLs to include the exploit q q For example, append exploit code instead of username Exploit code downloads the actual infection, joins the infected machine to a botnet, etc. 27

My. Doom q q q [Provos et al. ] Spreads by email My. Doom: searches local hard drive for addresses My. Doom. O: uses Web search engines q Queries split between Google (45%), Lycos (22. 5%), Yahoo (20%) and Altavista (12. 5%) Google’s view of My. Doom Peak scan rate: 30, 000 queries per second 28 Number of IP addresses generating queries (60, 000 hosts infected in 8 hours) Number of served queries drops as Google’s anomaly detection kicks in

Santy q Written in Perl, exploits a bug in php. BB bulletin board system (prior to version 2. 0. 11) q q [Provos et al. ] Allows injection of arbitrary code into Web server running php. BB Uses Google to find sites running php. BB Once injected, downloads actual worm code from a central site, asks Google for more targets and connects infected machine to an IRC botnet Multiple variants of the same worm q Polymorphism: actual Perl code changes from infection to infection, so filtering worm traffic is difficult! 29

Evading Anomaly Detection [Provos et al. ] q q q Google will refuse worm-generated queries Different Santy variants generate different search terms or take them from an IRC botmaster Google’s solution: if an IP address generates a lot of “rare” queries, ask it to solve a CAPTCHA 30

Index-Based Filtering [Provos et al. ] q q Idea: if worm relies on search results to spread, don’t provide vulnerable targets in search results During crawl phase, tag all pages that seem to contain vulnerable software or sensitive information such as email addresses q q Can’t drop them from the index because they may contain information useful to legitimate searchers Do not return the result of a query if it contains (a) pages from many hosts, and (b) high percentage of them are tagged as vulnerable q What are the limitations of this approach? 31

Witty Worm q q March 19, 2004, exploiting buffer overflow in firewall (ISS) products Infected 12, 000 machines in 45 mins Figures from “The Spread of the Witty Worm”, CAIDA 32

Witty Worm q First widely propagated worm w. destructive payload q q Seeded with more ground-zero hosts q q q 110 infected machines in first 10 seconds Shortest interval btw vulnerability disclosure & worm release q q Corrupted hard disk 1 day Demonstrate worms effective for niche too Security devices can open doors to attacks q q Other examples: Anti-virus software, IDS Installing security software does not mean more secure 33

Storm Worm / Peacomm (2007) q Spreads by cleverly designed spam campaign q Arrives as an email with catchy subject q q q Attachment or URL with malicious payload q q q First instance: “ 230 dead as storm batters Europe” Other examples: “Condoleeza Rice has kicked German Chancellor”, “Radical Muslim drinking enemies’s blood”, “Saddam Hussein alive!”, “Fidel Castro dead”, etc. Full. Video. exe, More. Here. exe, Read. More. exe, etc. Also masquerades as flash postcards Once opened, installs Trojan (wincom 32) & rootkit 34

Storm Worm Characteristics [Porras et al. ] q Infected machine joins botnet q q Obfuscated peer-to-peer control structure q q q Between 1 and 5 million machines infected (Sep 2007) Not like Agobot, which uses simple IRC control channel Interacts with peers via e. Donkey protocol Obfuscated code, anti-debugging defenses q q Goes into infinite loop if detects VMware or Virtual PC Large number of spurious probes (evidence of external analysis) triggers distributed Do. S attack 35

Storm Worm Outbreaks q Spambot binary used to spread new infections in subsequent campaigns q Looks for email addresses and mailing lists in the files on the infected machines 36