This year's BlackHat had a particularly wide range of topics. A more diverse range of topics means that more targets are under attack. This should come as no surprise.
On the one hand companies like Microsoft and Google have hardened their software against easy vulnerability exploitation.
On the other hand we're seeing a plethora of new (types of) devices getting equipped with network connectivity. Those devices come with limited or no built-in security.
This focus on embedded devices is also reflected in the amount of research done on the embedded devices within our phones and personal computers. There was a demonstration of a practical attack against smartphone baseband processors over the OMA DM protocol. The lack of built-in security, or the ability to easily add security controls, again demonstrates this major weakness in today's defense capabilities.
My colleague Vitaly Kamluk demonstrated, together with Anibal Sacco from Cubica Labs, how Computrace, popular low-jacking software, can be leveraged by attackers to perform remote code execution. Short of disabling the feature, which sometimes isn't even possible, there's no easy mitigation.
We're increasingly relying on more technology that can only be somewhat reliably protected by turning it off. This is not a sustainable path. While I appreciate that our personal operating systems appear to be getting safer, I'm rather pessimistic for the post-PC world.
Blackhat2014 USA was held at the Mandalay Bay, Las Vegas, after being held at Caesar's for the past 15 years or so. While the location wasn't central like Caesar's, the hotel was more spacious, easier to navigate, cleaner, and generally an improvement.
Jeff Moss inspired the Blackhat 2014 crowd with his introduction, and suggested that we are in a bubble that will pass. The cybersecurity crowd not only is finally taken seriously, but has never been more relevant, and is close to its peak of relevance. Access control manufacturers, medical device makers, automotive producers like Tesla, all of these enterprises are now not only listening. They are baking security better into their production processes, and this is good. We have made some progress. But with this relevance, when we look back, what can we say we contributed?
Dan Geer's rich keynote wound through a long list of policy proposals for a set of current cybersecurity challenges. Although I can't agree with everything he says, he is one of my favorite thinker-speaker-writers in the space and I find myself seeking out his talks. Interesting topics included mandatory reporting on cybersecurity failures on a national level, strikeback, resiliency, and the right to be forgotten. There were more. Many more. A text version of "Cybersecurity as Realpolitik" is online, and I recommend setting aside a block of time for dense, thoughtful material like this.
Dan supports mandatory reporting for breaches as laid out in "Surviving on a Diet of Poisoned Fruit", and real support for a cybersecurity failure reporting regime. He laid out current precedent in relevant law requiring reporting for other subject matter, and basic examples of why it would work. The CDC's communicable disease reporting and Mitre's anonymized near-miss aviation reporting center are strong examples of systems where self reporting on sensitive events, communicable disease and airline flight near misses, is both working and beneficial. He made the strong case that such a collection of cybersecurity data can be practical and is highly valuable, and it is time for national government or quasi-governmental entities to support transparent, anonymous collections.
The legitimacy of defensive strikeback is something that Dan dismissed onstage for strange reasons, but his text version gives it a LIMITED YES. He claimed that while Microsft and the Fbi are pulling off examples of successful strikeback activity, smaller organizations do not have the means to properly perform a strikeback effort. This is strange because small businesses can pool resources when they are under attack, co-ops and other pooling structures can help solve that problem. The US Chamber of Commerce, for example, is an incredibly powerful organization partly acting on behalf of small businesses pooling resources to get things done. The Chamber is incredibly powerful in the US, and small businesses benefit in some ways from some of those efforts (and perhaps suffer from some of them as well). And he didn't address the idea that strikeback doesn't need to be messy, extremely destructive, or sustained to put a quick end to a variety of attacks. But remotely stopping a large scale DDoS attack launched from compromised servers does not necessarily require attribution, and certainly does not require very messy or destructive solutions. They can be precise, and attribution is often possible. Everyone makes mistakes. He mentioned that difficulty in attribution compounds the problem. The bigger issue here is that these events cross multiple jurisdictions, definitions of law, and law enforcement organizations. And for some activity that would be otherwise criminal in different parts of the world, there is little if any enforcement.
Dan's take on the right to be forgotten brings out some unsettling Foucaltian subject matter, "I've spoken elsewhere about how we are all intelligence agents now, collecting on each other on behalf of various overlords" and that he wants to maintain some rights in the digital world to defend privacy, "I want to choose whether to misrepresent myself". But it's refreshing to hear a thought leader describe his desire and right to hop out of the digital panopticon.
Kaspersky Lab's Vitaly Kamluk, Sergey Belov, and Cubica's Anibal Sacco presented on "Absolute Backdoor Revisted", a surprising update on the state of security for a widely deployed low level utility. They presented a set of unusual behaviors implemented by this software that I rarely see exhibited by legitimate software but very frequently by malware. And then they abused it in multiple live demos, remotely wiping a brand new Windows 8 x64 laptop just out of the box via Absolute Computrace vulnerabilities.
Several years would seem to be ample opportunity to fix severe problems in update mechanisms, and it seems that the work has recently begun in response to the research review.
Web services came under serious attack this year. "Pivoting in Amazon Clouds" from Andres Riancho was interesting in light of the recent attacks on Elasticsearch sites abused in the Amazon cloud, demonstrating post-exploitation issues that enable attackers to further burrow into the cloud. He released a nice Nimbostratus tool and Hacme environment that aided his past penetration tests and EC2 security issue understanding. And Mobile Device Management systems were demonstrated to be chock full of web application problems with SQLi, poorly chosen crypto and encoding schemes, and poorly designed custom authentication schemes briefly put on the screen by Stephen Breen in "Mobile Device Mismanagement".
Promises to marry a potential groom if he covers his bride-to-be's travel expenses to his hometown are a fairly common feature of fraudulent spam. Less common are more 'noble' offers of help, though even these charitable offers usually come at a price.
In a recent mailing, a resident from the Ukrainian city of Odessa expressed his wish to become an organ donor, but for a considerable fee. In the email, he provided an overview of his current state of health – "good, not perfect" – his biometric data (height – 1.74 m, weight - 63 kg) and even his blood type. The price for which he was willing to sell a kidney or his liver was not specified; the main condition was that the operation had to be done in a European clinic.
It is obvious that anyone who decides to take him up on his offer will have to pay a considerable sum of money. It is highly likely that our "man from Odessa" will also want money to pay for his trip to Europe or to carry out tests in a good laboratory, before disappearing once he receives a money transfer. Honesty and offers made in spam are just incompatible. Moreover, no one should ever enter into negotiations with people who send unsolicited emails, especially when it concerns health issues.
Summer, sand and sun may let you think about vacation at a beach, but not IT-Security interested people. Every year gatherings happen in Las Vegas attracting amateurs and professionals from around the globe – BlackHat and Defcon. But also the local BSides conference, BSidesLV, takes place before - which teamed-up with Passwordscon this year.
Seven streams of presentations with a wide spectrum of topics were offered at Tuscany. At Passwordscon talks were given on securing passwords and attacking. Among these topics as "Target specific automated dictionary generation" which covered ways to automatically create dictionaries used for attacks against password hashes from one specified source. Rick Redman, from KoreLogic, gave a defense talk "Password Topology Histogram Wear-Leveling, a.k.a. PathWell". As attacks are getting easier because of better Hardware, not using stronger hash-types and defenses as password policies may lead to predictable passwords a new approach on defending passwords on enterprise level was presented. This is based on targeting the topology of passwords by limiting the use of password topologies and ban common password topologies.
Password topology means how the password is created. For example the topology of "Passw0rd" is uppercase (u) + 4x lowercase (l) + digit (d) + 2x lowercase (l) (simple: ulllldll). By using Levenshtein distance algorithm the change of topologies may be measured on a password change, to enforce new topologies rather than just updating any character in a password, to make cracking more difficult. [Link]
Dimitri Fousekis also focused on passwords in enterprise underlining the importance of "associate the password with data ownership" in order to avoid users disrespecting the importance of a good password.
Over the last 10 months, Kaspersky Lab researchers have analyzed a massive cyber-espionage operation which we call "Epic Turla". The attackers behind Epic Turla have infected several hundred computers in more than 45 countries, including government institutions, embassies, military, education, research and pharmaceutical companies.
The attacks are known to have used at least two zero-day exploits:
- CVE-2013-5065 - Privilege escalation vulnerability in Windows XP and Windows 2003
- CVE-2013-3346 - Arbitrary code-execution vulnerability in Adobe Reader
We also observed exploits against older (patched) vulnerabilities, social engineering techniques and watering hole strategies in these attacks. The primary backdoor used in the Epic attacks is also known as "WorldCupSec", "TadjMakhal", "Wipbot" or "Tavdig".
When G-Data published on Turla/Uroburos back in February, several questions remained unanswered. One big unknown was the infection vector for Turla (aka Snake or Uroburos). Our analysis indicates that victims are infected via a sophisticated multi-stage attack, which begins with the Epic Turla. In time, as the attackers gain confidence, this is upgraded to more sophisticated backdoors, such as the Carbon/Cobra system. Sometimes, both backdoors are run in tandem, and used to "rescue" each other if communications are lost with one of the backdoors.
Once the attackers obtain the necessary credentials without the victim noticing, they deploy the rootkit and other extreme persistence mechanisms.
The attacks are still ongoing as of July 2014, actively targeting users in Europe and the Middle East.
Note: A full analysis of the Epic attacks is available to the Kaspersky Intelligent Services subscribers. Contact: firstname.lastname@example.orgThe Epic Turla attacks
The attacks in this campaign fall into several different categories depending on the vector used in the initial compromise:
- Spearphishing e-mails with Adobe PDF exploits (CVE-2013-3346 + CVE-2013-5065)
- Social engineering to trick the user into running malware installers with ".SCR" extension, sometimes packed with RAR
- Watering hole attacks using Java exploits (CVE-2012-1723), Flash exploits (unknown) or Internet Explorer 6,7,8 exploits (unknown)
- Watering hole attacks that rely on social engineering to trick the user into running fake "Flash Player" malware installers
The attackers use both direct spearphishing and watering hole attacks to infect their victims. Watering holes (waterholes) are websites of interest to the victims that have been compromised by the attackers and injected to serve malicious code.
So far we haven't been able to locate any e-mail used against the victims, only the attachments. The PDF attachments do not show any "lure" to the victim when opened, however, the SCR packages sometime show a clean PDF upon successful installation.
Some of known attachment names used in the spearphishing attacks are:
- ؤتمر جنيف.rar (translation from Arabic: "Geneva conference.rar")
- NATO position on Syria.scr
- Talking Points.scr
- Security protocol.scr
In some cases, these filenames can provide clues about the type of victims the attackers are targeting.The watering hole attacks
Currently, the Epic attackers run a vast network of watering holes that target visitors with surgical precision.
Some of the injected websites include:
In total, we observed more than 100 injected websites. Currently, the largest number of injected sites is in Romania.
Here's a statistic on the injected websites:
The distribution is obviously not random, and it reflects some of the interests of the attackers. For instance, in Romania many of the infected sites are in the Mures region, while many of the Spanish infected sites belong to local governments (City Hall).
Most of the infected sites use the TYPO3 CMS (see: http://typo3.org/), which could indicate the attackers are abusing a specific vulnerability in this publishing platform.
The script "sitenavigatoin.js" is a Pinlady-style browser and plugin detection script, which in turn, redirects to a PHP script sometimes called main.php or wreq.php. Sometimes, the attackers register the .JPG extension with the PHP handler on the server, using "JPG" files to run PHP scripts:
The main exploitation script "wreq.php", "main.php" or "main.jpg" performs a numbers of tasks. We have located several versions of this script which attempt various exploitation mechanisms.
One version of this script attempts to exploit Internet Explorer versions 6, 7 and 8:
Unfortunately, the Internet Explorer exploits have not yet been retrieved.
Another more recent version attempts to exploit Oracle Sun Java and Adobe Flash Player:
Although the Flash Player exploits couldn't be retrieved, we did manage to obtain the Java exploits:Name MD5 allj.html 536eca0defc14eff0a38b64c74e03c79 allj.jar f41077c4734ef27dec41c89223136cf8 allj64.html 15060a4b998d8e288589d31ccd230f86 allj64.jar e481f5ea90d684e5986e70e6338539b4 lstj.jar 21cbc17b28126b88b954b3b123958b46 lstj.html acae4a875cd160c015adfdea57bd62c4
The Java files exploit a popular vulnerability, CVE-2012-1723, in various configurations.
The payload dropped by these Java exploits is the following:MD5: d7ca9cf72753df7392bfeea834bcf992
The Java exploit use a special loader that attempts to inject the final Epic backdoor payload into explorer.exe. The backdoor extracted from the Java exploits has the following C&C hardcoded inside:www.arshinmalalan[.]com/themes/v6/templates/css/in.php
This C&C is still online at the moment although it redirects to a currently suspended page at "hxxp://busandcoachdirectory.com[.]au". For a full list of C&C servers, please see the Appendix.
The Epic Turla attackers are extremely dynamic in using exploits or different methods depending on what is available at the moment. Most recently, we observed them using yet another technique coupled with watering hole attacks. This takes advantage of social engineering to trick the user into running a fake Flash Player (MD5: 030f5fdb78bfc1ce7b459d3cc2cf1877):
In at least one case, they tried to trick the user into downloading and running a fake Microsoft Security Essentials app (MD5: 89b0f1a3a667e5cd43f5670e12dba411):
The fake application is signed by a valid digital certificate from Sysprint AG:
Serial number: 00 c0 a3 9e 33 ec 8b ea 47 72 de 4b dc b7 49 bb 95
Thumbprint: 24 21 58 64 f1 28 97 2b 26 22 17 2d ee 62 82 46 07 99 ca 46
This file was distributed from the Ministry of Foreign Affairs of Tajikistan's website, at "hxxp://mfa[.]tj/upload/security.php".
The file is a .NET application that contains an encrypted resource. This drops the malicious file with the MD5 7731d42b043865559258464fe1c98513.
This is an Epic backdoor which connects to the following C&Cs, with a generic internal ID of 1156fd22-3443-4344-c4ffff:hxxp://homaxcompany[.]com/components/com_sitemap/
A full list with all the C&C server URLs that we recovered from the samples can be found in the technical Appendix.The Epic command-and-control infrastructure
The Epic backdoors are commanded by a huge network of hacked servers that deliver command and control functionality.
The huge network commanded by the Epic Turla attackers serves multiple purposes. For instance, the motherships function as both exploitation sites and command and control panels for the malware.
Here's how the big picture looks like:
The first level of command and control proxies generally talk to a second level of proxies, which in turn, talk to the "mothership" server. The mothership server is generally a VPS, which runs the Control panel software used to interact with the victims. The attackers operate the mothership using a network of proxies and VPN servers for anonymity reasons. The mothership also work as the exploitation servers used in the watering hole attacks, delivering Java, IE or fake applications to the victim.
We were able to get a copy of one of the motherships, which provided some insight into the operation.
It runs a control panel which is password protected:
Epic mothership control panel login
Once logged into the Control panel, the attackers can see a general overview of the system including the number of interesting potential targets:
A very interesting file on the servers is task.css, where the attackers define the IP ranges they are interested in. To change the file, they are using the "Task editor" from the menu. Depending on the "tasks", they will decide whether to infect the visitors or not. In this case, we found they targeted two ranges belonging to:
- "Country A" - Federal Government Network
- "Country B" - Government Telecommunications and Informatics Services Network
It should be noted though, the fact that the attackers were targeting these ranges doesn't necessarily mean they also got infected. Some other unknown IPs were also observed in the targeting schedules.
There is also an "except.css" file where attackers log the reasons they didn't try to exploit certain visitors. There are three possible values:
- DON'T TRY -> Version of the browser and OS does not meet the conditions
- DON'T TRY -> (2012-09-19 10:02:04) - checktime
These are the "don't meet the conditions" reasons observed in the logs:
- Windows 7 or 2008 R2
- MSIE 8.0
- Mozilla/4.0 (compatible; MSIE 8.0; Windows NT 6.1; WOW64; Trident/4.0; SLCC2; .NET CLR 2.0.50727; .NET CLR 3.5.30729; .NET CLR 3.0.30729; .NET CLR 1.1.4322; .NET4.0C; .NET4.0E)
- Adobe Shockwave 188.8.131.521
- Adobe Flash 10.3.181.14
- Adobe Reader 10.1.0.0
- Win Media Player 12.0.7601.17514
- Quick Time null
- MS Word null
- Java null
For this first stage of the attack, the threat actor uses a custom backdoor. In some cases, the backdoor is packaged together with the CVE-2013-5065 EoP exploit and heavily obfuscated. This makes the analysis more difficult.
The CVE-2013-5065 exploit allows the backdoor to achieve administrator privileges on the system and run unrestricted. This exploit only works on unpatched Microsoft Windows XP systems.
Other known detection names for the backdoor is Trojan.Wipbot (Symantec) or Tavdig.
The main backdoor is about 60KB in size and implements a C&C protocol on top of normal HTTP requests. The communication protocol uses requests in the C&C replies, which the malware decrypts and processes. The replies are sent back to the C&C through the same channel.
The malware behavior is defined by a configuration block. The configuration block usually contains two hard-coded C&C URLs. He have also seen one case where the configuration block contains just one URL. The configuration can also be updated on the fly by the attackers, via the C&C.
The backdoor attempts to identify the following processes and, if found, it will terminate itself:
It contains an internal unique ID, which is used to identify the victim to the C&C. Most samples, especially old ones, have the ID 1156fd22-3443-4344-c4ffff. Once a victim is confirmed as "interesting", the attackers upload another Epic backdoor which has a unique ID used to control this specific victim.
During the first C&C call, the backdoor sends a pack with the victim's system information. All further information sent to the C&C is encrypted with a public key framework, making decryption impossible. The commands from the C&C are encrypted in a simpler manner and can be decrypted if intercepted because the secret key is hardcoded in the malware.
Through monitoring, we were able to capture a large amount of commands sent to the victims by the attackers, providing an unique view into this operation. Here's a look at one of the encrypted server replies:
Once a victim is infected and "checks in" with the server, the attackers send a template of commands:
Next, the attackers try to move through the victim's network using pre-defined or collected passwords:
Listing all .doc files recursively is also a common "theme":
In total, we have decoded several hundreds of these command packages delivered to the victims, providing an unique insight into the inner workings of the attackers.
In addition to generic searches, some very specific lookups have been observed as well. These include searches for:
- eu energy dialogue*.*
In this case, the attackers were interested to find e-mails related to "NATO", "Energy Dialogue within European Union" and so on.
For some of the C&C servers, the attackers implemented RSA encryption for the C&C logs, which makes it impossible to decrypt them. This scheme was implemented in April 2014.Lateral movement and upgrade to more sophisticated backdoors
Once a victim is compromised, the attackers upload several tools that are used for lateral movement.
One such tool observed in the attacks and saved as "C:\Documents and Settings\All users\Start Menu\Programs\Startup\winsvclg.exe" is:Name: winsvclg.exe
Compiled: Tue Oct 02 13:51:50 2012
This is a keylogger tool that creates %temp%\~DFD3O8.tmp. Note: the filename can change across victims. On one Central Asian government's Ministry of Foreign Affairs victim system, the filename used was "adobe32updt.exe".
In addition to these custom tools, we observed the usage of standard administration utilities. For instance, another tool often uploaded by the attackers to the victim's machine is "winrs.exe":Name: winrs.exe
This is an UPX packed binary, which contains the genuine "dnsquery.exe" tool from Microsoft, unpacked MD5: c0c03b71684eb0545ef9182f5f9928ca.
In several cases, an interesting update has been observed -- a malware from a different, yet related family.Size: 275,968 bytes
Compiled: Thu Nov 08 11:05:35 2012
another example:Size: 218,112 bytes
Compiled: Thu Nov 08 11:04:39 2012
This backdoor is more sophisticated and belongs to the next level of cyber-espionage tools called the "Carbon system" or Cobra by the Turla attackers. Several plugins for the "Carbon system" are known to exist.
Note: the command and control servers www.losguayaberos[.]com and thebesttothbrushes[.]com have been sinkholed by Kaspersky Lab.
Other packages delivered to the victims include:MD5: c7617251d523f3bc4189d53df1985ca9
These top level packages deploy both updated Epic backdoors and Turla Carbon system backdoors to confirmed victims, effectively linking the Epic and Turla Carbon operations together.
The Turla Carbon dropper from these packages has the following properties:MD5: cb1b68d9971c2353c2d6a8119c49b51f
This is called internally by the authors "Carbon System", part of the "Cobra" project, as it can be seen from the debug path inside:
This acts as a dropper for the following modules, both 32 and 64 bit:MD5 Resource number 4c1017de62ea4788c7c8058a8f825a2d 101 43e896ede6fe025ee90f7f27c6d376a4 102 e6d1dcc6c2601e592f2b03f35b06fa8f 104 554450c1ecb925693fedbb9e56702646 105 df230db9bddf200b24d8744ad84d80e8 161 91a5594343b47462ebd6266a9c40abbe 162 244505129d96be57134cb00f27d4359c 164 4ae7e6011b550372d2a73ab3b4d67096 165
The Carbon system is in essence an extensible platform, very similar to other attack platforms such as the Tilded platform or the Flame platform. The plugins for the Carbon system can be easily recognized as they always feature at least two exports named:
Several Epic backdoors appear to have been designed to work as Carbon system plugins as well - they require a specialized loader to start in victim systems that do not have the Carbon system deployed.
Some modules have artifacts which indicate the Carbon system is already at version 3.x, although the exact Carbon system version is very rarely seen in samples:
The author of the Carbon module above can be also seen in the code, as "gilg", which also authored several other Turla modules.
We are planning to cover the Turla Carbon system with more details in a future report.
The payload recovered from one of the mothership servers (at newsforum.servehttp[.]com/wordpress/wp-includes/css/img/upload.php, MD5: 4dc22c1695d1f275c3b6e503a1b171f5, Compiled: Thu Sep 06 14:09:55 2012) contains two modules, a loader/injector and a backdoor. Internally, the backdoor is named "Zagruzchik.dll":
The word "Zagruzchik" means "boot loader" in Russian.
The Control panel for the Epic motherships also sets the language to codepage "1251":
Codepage 1251 is commonly used to render Cyrillic characters.
There are other indications that the attackers are not native English language speakers:
- Password it´s wrong!
- Count successful more MAX
- File is not exists
- File is exists for edit
The sample e9580b6b13822090db018c320e80865f that was delivered to several Epic victims as an upgraded backdoor, has the compilation code page language set to "LANG_RUSSIAN".
The threat actor behind the "Epic" operation uses mainly hacked servers to host their proxies. The hacked servers are controlled through the use of a PHP webshell. This shell is password protected; the password is checked against an MD5 hash:
The MD5 "af3e8be26c63c4dd066935629cf9bac8" has been solved by Kaspersky Lab as the password "kenpachi". In February 2014 we observed the Miniduke threat actor using the same backdoor on their hacked servers, although using a much stronger password.
Once again, it is also interesting to point out the usage of Codepage 1251 in the webshell, which is used to render Cyrillic characters.
There appears to be several links between Turla and Miniduke, but we will leave that for a future blogpost.Victim statistics
On some of the C&C servers used in the Epic attacks, we were able to identify detailed victim statistics, which were saved for debugging purposes by the attackers.
This is the country distribution for the top 20 affected countries by victim's IP:
According to the public information available for the victims' IPs, targets of "Epic" belong to the following categories:
- Ministry of interior (EU country)
- Ministry of trade and commerce (EU country)
- Ministry of foreign/external affairs (Asian country, EU country)
- Intelligence (Middle East, EU Country)
- Military (EU country)
- Research (Middle East)
- Pharmaceutical companies
- Unknown (impossible to determine based on IP/existing data)
When G-Data published their Turla paper, there were few details publicly available on how victims get infected with this malware campaign. Our analysis indicates this is a sophisticated multi-stage infection; which begins with Epic Turla. This is used to gain a foothold and validate the high profile victim. If the victim is interesting, they get upgraded to the Turla Carbon system.
Most recently, we observed this attack against a Kaspersky Lab user on August 5, 2014, indicating the operation remains fresh and ongoing.
Note: A full analysis of the Epic attacks is available to the Kaspersky Intelligent Services customers. Contact: email@example.com
We would like to add the following at the end of the blogpost, right before the detection names:
If you'd like to read more about Turla/Uroburos, here's a few recommendations:
- G-Data's paper "Uroburos Highly complex espionage software with Russian roots"
- BAE Systems analysis of "The Snake campaign"
- "Uroburos: the snake rootkit", technical analysis by deresz and tecamac
- "TR-25 Analysis - Turla / Pfinet / Snake/ Uroburos" by CIRCL.LU
Kaspersky products' detection names for all the malware samples described in this post: