End-to-End Analysis of a Domain Generating Algorithm Malware Family
3 likes9,378 views
This document summarizes the analysis of a domain generating algorithm (DGA) malware family. Key points include: - The malware uses inline code obfuscation and encrypted strings to hide its functionality and communication domains. Researchers were able to deobfuscate the code and decrypt the strings to analyze the malware. - Clues in decrypted strings suggest the malware author is Romani, including references to Romani singers in template strings. - The malware generates domain names by concatenating two randomly selected words from a dictionary and appending ".net". This allows it to generate many domain variations to communicate with its command and control servers. - The DGA algorithm uses a 15-bit seed value derived from the
![Generation of One-Time Pad
for
(i
=
0;
i
<
lengthOfOneTimePad;
i
+=
4)
{
oneTimePad[i
+
0]
=
(seed
>>
0x00)
&
0xFF;
oneTimePad[i
+
1]
=
(seed
>>
0x08)
&
0xFF;
oneTimePad[i
+
2]
=
(seed
>>
0x10)
&
0xFF;
oneTimePad[i
+
3]
=
(seed
>>
0x18)
&
0xFF;
seedRotated
=
((seed
>>
1)
|
(seed
<<
(32
-‐
1)));
seed
=
(seedRotated
&
0xFFFF0000)
|
((seedRotated
+
((seedRotated
>>
0x08)
&
0xFF))
&
0xFF)
<<
0x08)
|
((2
*
seedRotated
+
((seedRotated
>>
0x08)
&
0xFF))
&
0xFF);
}](https://image.slidesharecdn.com/end-to-endanalysisofadomaingeneratingalgorithmmalwarefamily-slides-130814103104-phpapp02/85/End-to-End-Analysis-of-a-Domain-Generating-Algorithm-Malware-Family-21-320.jpg)
![Statically Decrypting All Strings
for
(i
=
0;
i
<
(lengthOfOneTimePad
–
0x0C);
i++)
{
beginningOfStrings[i]
^=
oneTimePad[0x0C
+
i];
}](https://image.slidesharecdn.com/end-to-endanalysisofadomaingeneratingalgorithmmalwarefamily-slides-130814103104-phpapp02/85/End-to-End-Analysis-of-a-Domain-Generating-Algorithm-Malware-Family-22-320.jpg)
![string
GetHostname(UInt32
seed)
{
byte[]
aShuffle
=
new
byte[15];
for
(int
i
=
0;
i
<
15;
i++)
{
aShuffle[aHelperTable[i
*
2]]
=
(byte)(seed
&
1);
seed
>>=
1;
}
int
iHost1
=
0;
int
iHost2
=
0;
for
(int
i
=
0;
i
<
7;
i++)
{
iHost1
=
2
*
iHost1
|
aShuffle[i];
iHost2
=
2
*
iHost2
|
aShuffle[i
+
7];
}
iHost2
=
(2
*
iHost2
|
aShuffle[14])
+
128;
UInt16
offsetHost1
=
(UInt16)((UInt16)(aHexHostname[iHost1
*
2])
+
(UInt16)(((UInt16)(aHexHostname[iHost1
*
2
+
1]))
<<
0x08));
UInt16
offsetHost2
=
(UInt16)((UInt16)(aHexHostname[iHost2
*
2])
+
(UInt16)(((UInt16)(aHexHostname[iHost2
*
2
+
1]))
<<
0x08));
string
host1
=
"";
string
host2
=
"";
byte
b;
while
((b
=
aHostStrings[offsetHost1++])
!=
0)
{
host1
+=
(char)b;
}
while
((b
=
aHostStrings[offsetHost2++])
!=
0)
{
host2
+=
(char)b;
}
return
host1
+
host2
+
".net";
}](https://image.slidesharecdn.com/end-to-endanalysisofadomaingeneratingalgorithmmalwarefamily-slides-130814103104-phpapp02/85/End-to-End-Analysis-of-a-Domain-Generating-Algorithm-Malware-Family-33-320.jpg)
End-to-End Analysis of a Domain Generating Algorithm Malware Family
- 1. End-to-End Analysis of a Domain Generating Algorithm Malware Family Jason Geffner Sr. Security Researcher CrowdStrike, Inc. Black Hat USA 2013
- 2. Domain Generating Algorithms • Most modern malware communicates with attackers’ servers • Typical bots/RATs/downloaders • DGA malware
- 3. DGA Example • Every minute, have malware connect to GMT-time- based server address • <month><day><year><hour><minute>.com • On July 31, 2013, at 2:30 PM, malware connects to 0731131430.com • Attacker registers domain and server prior to strike-time • Attacker redirects domain and takes down server immediately aer strike-time
- 4. Notable DGA History • Early 2008 – Kraken one of the first malware families to use a DGA • Late 2008 – Conficker first discovered • 2010 – Texas A&M University researchers publish paper on detecting DGA domain names • 2012 – Damballa releases whitepaper on new DGA use in six malware families
- 5. New DGA Family • In February of 2013, a major American financial services firm received a suspicious email with an EXE file attachment • Firm’s CISO sent the attachment to their “global cyber intelligence” partner, who had trouble analyzing it: “It is the obfuscation that is throwing redacted off.” • As a result, the CISO forwarded it to us REDACTED
- 6. Code Obfuscation • Most obfuscated malware is obfuscated with a packer • This malware uses inline code obfuscation
- 7. This snippet of disassembly shows random 32-bit values being assigned to variables and used in mathematical calculations. This junk code is interspersed with legitimate code.
- 8. x ✔ x x x x x x x x x x ✔ x x x ✔ ✔ x x x x This snippet of disassembly shows random 32-bit values being assigned to variables and used in mathematical calculations. This junk code is interspersed with legitimate code.
- 12. Code Deobfuscation • Find all basic legitimate variables • Function arguments to the current function • Global variables • Local function variables used as parameters to function calls • Local function variables that store return values of function calls • All other local function variables considered legitimate if their values are read from or written to other legitimate variables
- 13. Decompilation Graph, Before Deobfuscation
- 14. Decompilation Graph, Aer Deobfuscation
- 15. Demo CrowdDetox
- 16. CrowdDetox • Free open-source plugin for Hex-Rays • Finds legitimate variables and code in a decompilation graph, and prunes everything else • Available at http://www.crowdstrike.com/community-tools
- 17. Data Obfuscation • EXE contains no readable static strings related to malicious functionality • No registry keys • No file names • No server addresses • No URI paths • All strings are decrypted at run time
- 19. Dynamically Deobfuscating Data • Within first hour of incident response • Found string decryption function • Identified list of encrypted strings • Patched binary to decrypt strings in-place as opposed to on heap • Patched binary with hand-written assembly to call string decryption function on each encrypted string
- 20. Statically Deobfuscating Data • String decryption function XORs encrypted strings with one-time pad • One-time pad is generated at run time
- 21. Generation of One-Time Pad for (i = 0; i < lengthOfOneTimePad; i += 4) { oneTimePad[i + 0] = (seed >> 0x00) & 0xFF; oneTimePad[i + 1] = (seed >> 0x08) & 0xFF; oneTimePad[i + 2] = (seed >> 0x10) & 0xFF; oneTimePad[i + 3] = (seed >> 0x18) & 0xFF; seedRotated = ((seed >> 1) | (seed << (32 -‐ 1))); seed = (seedRotated & 0xFFFF0000) | ((seedRotated + ((seedRotated >> 0x08) & 0xFF)) & 0xFF) << 0x08) | ((2 * seedRotated + ((seedRotated >> 0x08) & 0xFF)) & 0xFF); }
- 22. Statically Decrypting All Strings for (i = 0; i < (lengthOfOneTimePad – 0x0C); i++) { beginningOfStrings[i] ^= oneTimePad[0x0C + i]; }
- 23. Malware Family Template • Initially collected over 100 variants from this malware family • Most use randomized strings in decrypted data • File names • Directory names • Registry names • A few use template placeholders instead of randomized strings
- 24. Malware Family History • CrowdStrike initially collected: • 16 “template” variants • 132 non-“template” variants • PE Timestamps appear to be legitimate • Malware first built and distributed in February, 2013
- 25. Variants Grouped by PE Timestamp 0 5 10 15 20 25 30 35 40 45 2/2/13, 21:00 2/2/13, 22:00 … 2/5/13, 18:00 2/5/13, 19:00 2/5/13, 20:00 "Template" Variants Non-"Template" Variants NumberofVariants PE Timestamp, Bucketed by Hour (UTC)
- 26. Authorship Clues in Decrypted Strings • All variants using default template have the same seed value: 0x445A4950 • Parsed as ASCII, 0x445A4950 = “PIZD” • “PIZD” translates from Bosnian / Croatian / Latvian / Polish / Romani / Romanian / Slovenian to English as “PUSSY”
- 27. Authorship Clues in Decrypted Strings • Template string for copied file name is “XZSEQWSpulaosugiingat.exe” • “pula o sug i în gât” loosely translates from Romanian to English as “suck a dick in your throat”
- 28. Authorship Clues in Decrypted Strings • However, a Romanian is more likely to say, “suge pula în gât” • “pula o sug i în gât” is more likely the wording a Romani would use • Additionally, a Romanian is more likely to say “pizda” than “pizd”; a Romani would say “pizd”
- 29. Decrypted Strings Suggest Romani Author • Template strings for directory name and registry value names are “NICOLAEGUTAXZSEQWS” and “COSTIIONITAEQWS” • Template string used in entry point obfuscation is “ADRIANCOPILUMINUNESIFLORINSALAM” • Nicolae Guţă, Costi Ioniță, Adrian Copilul Minune, and Florin Salam are prominent Romani manele singers
- 30. Domain Generating Algorithm • All variants of family contain identical 384-word list of common English words, decrypted at run time • Domain names created by concatenating two pseudo-randomly selected words and appending “.net” to the end
- 32. Domain Generating Algorithm • Pseudo-random algorithm uses only 15 bits of the seed value, so only 32,768 possible domain names Seed Value = seconds elapsed since January 1, 1970 GMT/512 ∴ Seed Granularity = 512 seconds = 8 minutes and 32 seconds • Malware tries 85 domains per seed value (seed+0, seed+1, seed +2, …), creating a sliding “window” of DGA domains
- 33. string GetHostname(UInt32 seed) { byte[] aShuffle = new byte[15]; for (int i = 0; i < 15; i++) { aShuffle[aHelperTable[i * 2]] = (byte)(seed & 1); seed >>= 1; } int iHost1 = 0; int iHost2 = 0; for (int i = 0; i < 7; i++) { iHost1 = 2 * iHost1 | aShuffle[i]; iHost2 = 2 * iHost2 | aShuffle[i + 7]; } iHost2 = (2 * iHost2 | aShuffle[14]) + 128; UInt16 offsetHost1 = (UInt16)((UInt16)(aHexHostname[iHost1 * 2]) + (UInt16)(((UInt16)(aHexHostname[iHost1 * 2 + 1])) << 0x08)); UInt16 offsetHost2 = (UInt16)((UInt16)(aHexHostname[iHost2 * 2]) + (UInt16)(((UInt16)(aHexHostname[iHost2 * 2 + 1])) << 0x08)); string host1 = ""; string host2 = ""; byte b; while ((b = aHostStrings[offsetHost1++]) != 0) { host1 += (char)b; } while ((b = aHostStrings[offsetHost2++]) != 0) { host2 += (char)b; } return host1 + host2 + ".net"; }
- 34. Malware’s Use of DGA • Malware regularly connects to DGA domains to send HTTP GET request • /forum/search.php?email=<hardcoded email address>&method=post • Each malware variant has a unique hardcoded email address
- 35. Malware’s Use of DGA • If the server’s response contains the correct fingerprint, the malware requests the same URL again • If the server’s second response contains the correct fingerprint, the malware saves the downloaded content as an EXE and executes it
- 36. Sinkholing • Sinkholed five domains to which the DGA would resolve on: • March 5, 2013 • March 6, 2013 • March 7, 2013 • March 8, 2013 • March 9, 2013 • Nearly 15,000 hits from malware • Logged 1,170 unique client IP addresses • Logged 1,000 unique email addresses posted to sinkhole servers
- 38. Example Email Addresses Posted to Sinkholes • 1800flowers@1800reminders.com • billing@deluxeforbusiness.com • consultant_fiscal-unsubscribe@yahoogroups.com • fbmessage+fepvdccz@facebookmail.com • geico_claims@geico.com • northwest.airlines@nwa.com
- 39. More Email Addresses Posted to Sinkholes • 421 personal yahoo.com addresses • 66 personal aol.com addresses • 59 personal hotmail.com addresses • 31 personal comcast.net addresses • 4 .gov addresses • 1 .mil address • 0 gmail.com addresses
- 40. Sinkhole Hits From IP Addresses 1. United States 2. Romania 3. Japan 4. Russia 5. Germany 6. France 7. India 8. Netherlands 9. United Kingdom 10. Sweden 575 321 46 17 15 15 14 14 13 11
- 41. DGA Monitoring • Developed automation solution to download from DGA domains in real-time • Domains connected to campaign responded with identifiable HTTP response fingerprint • Automation ran for two weeks • Detected 20 domains connected to campaign
- 42. Domain Analysis • 19 of the 20 domains • Registered via and hosted by Yahoo! Inc.’s Small Business hosting plan • Registrants used @yahoo.com email account • 1 of the 20 domains • Registered via and hosted by Omnis Network LLC • Registrants used @aol.com email account
- 43. Domain Analysis • All domains registered 0-48 hours before DGA pointed to them • Identical registrant names and addresses used for several domains, with semi-random phone numbers corresponding to city area code
- 44. DOMAIN REGISTRANT ADMIN EMAIL ADMIN PHONE amountcondition.net Robert Seifert 2212 W. Farwell Chicago, IL 60645 seifertrobertw@yahoo.com +1.7737916544 weathereearly.net Robert Seifert 2212 W. Farwell Chicago, IL 60645 robertwseifert@yahoo.com +1.7737916324 heardstrong.net Lynette Conlan 210 Pinehurst Way San francisco, CA 94080 donnybonham184@yahoo.com +1.6505882763 variousopinion.net Lynette Conlan 210 Pinehurst Way San francisco, CA 94080 alankimberley@yahoo.com +1.6505882742 morningpaint.net clint Bertke 299 lowry rd fort recovery, OH 45846 clintmbertke@yahoo.com +1.4198523054 withinshould.net bertke, clint m 299 lowry rd fort recovery, OH 45846 clintmbertke@aol.com +1.4198523054
- 45. Historic WHOIS Research • 7 of the 20 domains hosted blank root webpages during WHOIS research • 3 of the 20 domains’ webservers were down during WHOIS research • The other 10 domains all hosted content for “GlobalPartners Hungaria K.”
- 48. Sokratis Kokkalis Chairman and CEO of Intracom Holdings
- 49. Petros Souretis Managing Director of INTRAKAT, a subsidiary of Intracom Holdings Sokratis Kokkalis Chairman and CEO of Intracom Holdings
- 50. Petros Souretis Managing Director of INTRAKAT, a subsidiary of Intracom Holdings Sokratis Kokkalis Chairman and CEO of Intracom Holdings INTRAKAT’s logo
- 51. Scanning All DGA Domains • Scanned root webpage of all 32,768 possible DGA domains for “GlobalPartners” • Found 44 additional domains, for a total of 64 campaign domains • All but two registered through a Yahoo! Small Business hosting plan • All domains registered for exactly one year • Oldest domain registered on February 3rd, 2013
- 52. Campaign Domain Registrant Email Addresses • Email addresses primarily fall into one of four categories 1. Related to name of domain’s registrant (marcosuriano21@yahoo.com for Marco Suriano) 2. Related to name of another domain’s registrant, likely a mistake made by adversary (ike2ricchio4@yahoo.com for Kai Roth) 3. Related to domain name (degreeanimal@yahoo.com for degreeanimal.net) 4. Related to domain name of another domain (degreeanimal@yahoo.com for nightwagon.net)
- 53. Non-DGA Domains Used in Campaign • Open-source research led to non-DGA domains also used in this campaign • antaragroup.org • ahai-group.com • azrhgroup.com • fastwire.us • int-group.us • international-wire.com • intracombusiness.com • intracomfinancial.com • itpservices.us • kpl-business.com • logicom-holding.com • mtkoffice.co.uk • rbs-partners.com • trust-core.net
- 55. Previously Researched Campaign History • Further investigation leads to research on anti-fraud site http://www.bobbear.co.uk/
- 56. Extended Campaign History • March 2013 • March 2013 • February 2013 • January 2013 • September 2012 • July 2011 • April 2011 • November 2009 Trust Core Mojo Directo GlobalPartners Anatara Group Ahai Group Azure Holding Group KPL Logicom • May 2009 • February 2009 • December 2008 • November 2008 • June 2008 • January 2008 • September 2007 • May 2007 RBS Partners FastWire Group INTRACOM MTK ITP International Wire INT Group Interpay Group
- 57. Antivirus Detections • Malware appears to have begun circulating in February 2013 • Our analysis conducted in February and early March of 2013 • Avast discovered a variant of it in June of 2013 - https://blog.avast.com/2013/06/18/your-facebook- connection-is-now-secured/
- 58. Antivirus Detections Detection Rate Engine Most Common Detection Detection Rate Engine Most Common Detection 100.0% Malwarebytes Trojan.Agent 88.4% Fortinet W32/Agent 99.3% ESET Win32/Agent 81.0% Norman Malware 98.6% AVG Generic_r 76.9% GFI VIPRE Trojan.Win32.Agent 98.6% Kaspersky Trojan.Win32.Generic 75.5% Avast Win32:Agent 98.0% Panda Trj/Genetic 38.1% McAfee Artemis 98.0% Sophos Troj/Agent 21.8% Trend Micro TROJ_GEN 95.2% G Data Gen:Variant.Zusy 17.7% Symantec WS.Reputation.1 93.2% Bitdefender Gen:Variant.Zusy 15.0% Microso Win32/Suppobox 91.8% F-Secure Gen:Variant.Zusy 0% ClamAV
- 59. Conclusion • DGA downloader likely authored by Romani male, who appears to be working with a long-running European money mule crime syndicate • Another component apparently harvests email addresses, builds the DGA component, and emails it to target recipients • DGA domains appear to be registered using stolen credit card numbers • Inlined code obfuscation can be defeated with new CrowdDetox plugin for Hex-Rays
- 60. Q & A
- 61. Special thanks to Alex Ionescu for Romanian translations
- 62. @CrowdStrike