Thursday, July 05, 2018

LNK "Toolmarks"

LNK Artifacts and "Toolmarks"
I've discussed LNK files in this blog a number of times over the years, mostly focusing on the file format.  In one instance (from 5 years ago), the focus was on parsers, and specifically about determining devices that had been connected to the system, as when it comes to illicit images/videos, such things can determine the difference between "possession" and "production".  This sort of information can also be used to address such topics as data exfiltration.

Last year, I started looking at LNK files (here, and then shortly thereafter, here...) from a different perspective; that of LNK files being sent as email attachments, or somehow being delivered to a "victim".

So, what?  Why does this matter?  Most users of Windows systems are very familiar with LNK files, or "Windows shortcuts", as they pertain to user actions on a system, or software installations.  Both of these generally result in an LNK file being created.  And if you follow malware write-ups or threat intel blog posts, you'll very likely see mention of some malware variant or targeted adversary creating an LNK file in a user's StartUp folder as a means of remaining persistent on the system.  These LNK files are usually created on the compromised system, through the use of the local API, so other than their location, there's nothing too terribly interesting about them.

However, when an adversary sends a Windows shortcut file as an email attachment (or through some other mechanism) that LNK file contains information about their (the adversary's) system.  The folks at JPCERT/CC mentioned this about a year and half ago (the folks at NViso Labs said something similar, as well); if the LNK file was created (via the MS API) on the adversary's system, then it likely contains the NetBIOS name of the system, the MAC address, the volume serial number, and the code page number (although in the research I've done so far, I have yet to find an LNK file with the code page number populated).  While anyone of these values may be seen as trivial, all of them together can be employed to great effect by researchers. For example, using the NetBIOS name, MAC address, and VSN (such as illustrated in this blog post), I wrote a Yara rule that I deployed as a VirusTotal retro-hunt, in order to look for other examples; I ended up with several dozen responses in very short order.  An example of such a rule is the "shellphish" Yara rule illustrated in this blog post.  In the case of the "shellphish" rule, the LNK file that I initially looked at also included a SID in the Property Store Data Block, and including it in the retro-hunt served to minimize false positives (i.e., hits on files that contained the other byte sequences, but were not what I was looking for).

Using an LNK file parser available online, I parsed a LNK file with an embedded, base64-encoded Powershell command, and saw the following:

[Distributed Link Tracker Properties]
Version: 0
NetBIOS name: RAA6AFwAYQBhAC4AĆ£\ J G
Droid volume identifier: 69426764-4841-414d-0000-000000000032
Droid file identifier: e09ff242-e6d1-9945-a75d-f512f84510dc
Birth droid volume identifier: 8e3691fc-e7be-bc11-6200-1fe2f6735632
Birth droid file identifier: e09ff242-e6d1-9945-a75d-f512f84510dc
MAC address: f5:12:f8:45:10:dc
UUID timestamp: 03/17/3700 (13:54:42.551) [UTC]
UUID sequence number: 10077
The file was moved to a new volume.

Okay, so, this is very different.  Usually, the "NetBIOS name" is 16 bytes, which is what my own parser looks for; however, in this case, what is the "machine ID" consumes a good bit more than 16 bytes, as illustrated in figure 1.

Fig 1: Hex dump from LNK file; Tracker Data Block








In this case, as you can see in figure 1, the "machine ID" is 31 bytes in size, with the first 24 bytes being "RAA6AFwAYQBhAC4AdgBiAHM", and the last 7 being all zeros.  As the size of the Tracker Data Block (4 bytes at 0xca3) remains consistent at 96 bytes, the larger-than-expected machine ID presents a parsing problem.  The "droids" start at offset 0xcd2, and are identical, so due to the larger machine ID, the Tracker Data Block itself "ends" part way through that section of data, at offset 0xd02.  This leaves the remaining 15 bytes of the final droid (which ends at 0xd11) unparsed...notice how the math works out?  The machine ID is 15 bytes larger than it should be, and as such, everything is "off" (per the MS file format documentation) by 15 bytes.

Normally, my parser returns output that appears as follows:

***TrackerDataBlock***
Machine ID            : admin-pc
New Droid ID Time     : Tue Sep 19 12:57:20 2017 UTC
New Droid ID Seq Num  : 11190
New Droid    Node ID  : 08:00:27:3a:ce:83
Birth Droid ID Time   : Tue Sep 19 12:57:20 2017 UTC
Birth Droid ID Seq Num: 11190
Birth Droid Node ID   : 08:00:27:3a:ce:83

The above output is from an LNK file that fully meets the file format specification.  During my research into some of these malicious LNK files, I added the ability to provide debugging information so that anomalies would be more apparent, rather than "hidden" behind the parsing process.

From the LNK file from which the hex dump in figure 1 was extracted, we can still extract and manually parse the droids; as the ObjectIDs are based on the UUID version 1 format (discussed in this blog post), we can use this information to our advantage in our research, extracting the node IDs (MAC addresses), etc.  Interestingly enough, though, the larger machine ID doesn't seem to have an effect on the command embedded within the LNK file; having sections of the file format that do not comply with the specification doesn't seem to affect the performance of the LNK file.

I began wondering if the larger machine ID field might be a "toolmark" of some kind, an artifact of whatever process or application that had been used to create the LNK file.  I also began looking at a range (I hesitate to qualify the range with "wide"...) of LNK files found online, particularly those that ran external commands (encoded Powershell commands, bitsadmin commands, etc.), as well as looking for tools, applications, or processes that would allow someone to create arbitrary, malicious LNK files.  At this point in time, I haven't yet found one that produces "toolmarks" similar to what's seen in figure 1.

The machine ID/NetBIOS name, user SID, VSN, and MAC address are all artifacts from the adversary's system that remain within the LNK file when it is sent to a target.  The folks at ProofPoint discussed one or two means of deploying LNK files, which could result in some very interesting artifacts.  I'm sure that if enough specifically-malicious LNK files were examined, we'd be able to pull out "toolmarks" from such things as specific applications/tools or processes used to create (or modify) the LNK files, or perhaps even identify misbehaving applications that create LNK files on non-Windows systems.

Friday, June 29, 2018

Updates

I just pushed three new plugins up to the RegRipper plugin repository, two written by Gabriele Zambelli (photos_win10.pl, msedge_win10.pl), and I wrote source_os.pl, to address the "When Windows Lies" issue that Mari brought up last year.

Addendum, 30 June: Pushed a new plugin from M. Godfrey, named "imgburn1.pl" to the repo this afternoon. Many thanks to Michael (and Gabriele) for writing and submitting plugins!

Addendum, 2 July: Thanks to input and test data from Mitch Impey, I was able to quickly update not only shellbags.pl and shellbags_tln.pl, but also comdlg32.pl.  Providing sample/test data makes troubleshooting and updating current plugins, or creating new ones, a much quicker and more efficient process.  Thanks, Mitch!

Addendum, 5 July: Many thinks to Micah Jones for sharing the dafupnp.pl plugin he wrote created!  This is a plugin that pulls information about media streaming devices from the System hive.  Thanks, Micah, for the great work and for sharing the plugin.  Also, based on the data that Micah shared, I updated the bthport.pl plugin, as well.

Also, I added bthport_tln.pl to the repository, as well.  This will help in performing timeline analysis for such things as data exfil via Bluetooth devices.

Thursday, June 21, 2018

More about EDR/MDR

On the heels of my previous post on the topic, from my perspective, there still seems to be something of a misconception within the market as a whole as to the value of EDR/MDR. 

During conversations on the EDR/MDR topic, one of the questions we hear quite often is, "Do you detect X?", with "X" being some newly-identified virus or exploit.  I think that the reason we hear this question so often is that EDR/MDR is still somewhat new in the minds of the marketplace, and most organizations are still trying to visualize and really understand the concept, and how it applies to them.  The way we most often go about doing that sort of thing is by finding something we're familiar with, and start by building a comparison relationship there.  Unfortunately, starting with AV as the comparison doesn't really get us to where we need to be, in part because this is a whole new way of looking at things. 

And believe me, I do understand how difficult it is in today's marketplace to really understand what's out there.  A while back, I was in a sales presentation, and a sales rep was going through the slide pack they'd been provided by marketing; at one point, the client just came right out and said, "...yes, everyone says that, every other vendor we've spoken to has said the same exact thing.  How are you different?"  They were asking this question because they were trying to get beyond the initial fluff and down to, okay, how is this vendor's product better for us and our needs? Now, the folks we were talking to were actually very mature, in the sense that they had already identified their EDR needs, and actually had been working with an EDR product with a subset of their infrastructure, so they were essentially trying to make an apples-to-apples comparison of products.

As recently as last week, I was encouraged via a marketing email to sign up for a webinar, and one of the inducement statements in the corresponding blog post (linked in the email) was along the lines of, "...you should be detecting ransomware the moment it reaches out to the C2 server."

Hhhhmmm...but what if we could detect the issue before that point?  What if you could detect the initial behaviors that lead to a malware infection, and block the activity at that point, rather than waiting to try to detect the malware itself?  If we focused our efforts earlier in the attack cycle, we could impose greater pain on the attacker, raising the level of effort required on their part, and forcing them to change their tactics.

So what do I mean by this?  Every day, thousands (millions?) of employees log into their corporate computer systems, launch Outlook, and read their email.  Part of this may involve opening attachments, including MSWord documents and Excel spreadsheets.  As such, it's normal and expected to see a process tree that goes from the user logging in (Windows Explorer), to launching Outlook, and then Outlook includes whichever attachment is opened as a child process.  It looks something like this:

explorer.exe
    |__Outlook.exe
              |__winword.exe

Now, when a user is sent a phishing email with a weaponized attachment, the user may be prompted to click the "Enable Content" button in MSWord in order to allow any embedded macros to run; once this happens, we generally see something like the command prompt (cmd.exe), or something else (PowerShell, etc.) running as a child process of MSWord.  This is ( a ) never a good thing, ( b ) easy to detect, and ( c ) equally trivial to block.  Here's what that might look like:

explorer.exe
    |__Outlook.exe
              |__winword.exe
                        |__cmd.exe /c powershell.exe...
                                |__powershell.exe...

Here's a good example, shared by Phil Burdette some time ago, albeit without the Outlook component. 

The point is that this behavior can be detected and alerted on, and then action can be take at the point where winword.exe spawns a command shell as a subprocess.  As such, if we're blocking processes from executing at that point, then we're not even getting to the point where PowerShell or WScript or BITSAdmin or whatever is used to download the malicious stuff to the compromised system.  With the right instrumentation, you may also be able to isolate the system on the network after having blocked the malicious behavior.  That way, instead of sending someone a ticket telling them that they have to respond, and them getting it at some point in the future (even as much as a few minutes is enough for the adversary to burrow their way into your infrastructure), all access is immediately disabled.

This is where MDR solutions come into play.  With the right endpoint technology in place, an MDR not only monitors what's going on, but is able to take threat intelligence developed from monitoring their entire client base, and apply is across all monitored clients.  This means that every client benefits from what was gleaned and developed from any one client, and they benefit immediately.

What this ultimately leads to is much earlier detection of malicious activity, and a much quicker response to that activity, such that reporting is obviated.  After all, if you're able to detect and respond to an adversary much earlier in their attack cycle, and you're able to cycle through the OODA loop faster than the adversary, then you're able to prevent access to any "sensitive data".  If that sensitive data isn't accessed, then you don't have to report.  Also, you've got a record of activity that demonstrates that you were able to respond to, contain, and eradicate the adversary.

Monday, June 18, 2018

Updates

I've had a bunch of draft blog posts sitting around, and for some of them, particularly the really technical ones, I thought, "...ah, no one's gonna read this...", so I opted for another medium, or I simply deleted them.  However, I decided to throw a couple of them in together, into one post, just so I could complete my thoughts and get them out there, and do a bit of housekeeping with respect to my blog drafts.

Publishing
Brett Shavers recently posted some interesting content on the topic of publishing in DFIR.  While DFIR doesn't necessary follow the "publish or perish" adage most often seen in academia, Brett does have some very good points.

To Brett's point about "lack of contributors", this is the reason why efforts like Into The Boxes project never really took off...you can't sustain a community-based project if folks within the community aren't willing to contribute.

By the way, I still have my inaugural hard copy of "Into The Boxes #1", knowing full well that one day it's going to be part of my DFIR museum, right alongside my MS-DOS install diskettes, my original Windows XP install CDs, and an AOL installation CD!

Something else that Brett mentioned in his post was "peer review".  Oddly enough, I've been engaged in a conversation with someone else in the DFIR community about this very topic, particularly as it applies to analysis findings.  In my experience, peer reviews within the community are spotty, at best.  We all know that team member that we can go to, because they'll turn around a 40 page report draft in 15 minutes, saying just that it "looks good."  And we also know that team member we can go to and rely on to catch stuff we may have missed.

Blog-a-Day Challenge
Speaking of publishing, it looks as if over the last couple of weeks a number of folks have decided to take the Zeltser Challenge, that is to post a blog a day for an entire year.  Not only has David Cowen picked it back up (he's done it before...), but others have joined the party, as well.  For example, Stacey has started The Knowledge Bean and decided to partake in the challenge. Tony at Archer Forensics has picked up the mantle, as well, and is off and running.

If anyone else in DFIR is doing something like this...blog-a-day, or even blog-a-week, let me know...I'd like to check it out.

Oh, and tying this back to the previous topic, something Tony mentioned in one of the blog posts:

I would solicit everyone to do that deeper dive research to further the field.

Agreed, 1000%. 

New Plugins
Well, let's see...what's new on the plugin front?  So, I recently received one updated plugin (lastloggedon.pl) and a new plugin (utorrent.pl) from Michael Godfrey (author information included in the plugins), and uploaded them to the repository.  I also added jumplistdata.pl, which is based on this finding from Costas K.  Finally, I created execpolicy.pl, a plugin to check the PowerShell ExecutionPolicy value, if it exists...some folks have said that adversaries have been observed modifying the Registry value to that they can bypass the execution policies in order to run their code, albeit not from the command line.

Not a new plugin, but I also updated the license for RegRipper to the MIT license, at the request of some folks at AWS.

Tool Usage
Mari published a blog article recently, which I thought was great...not just because Mari's great at sharing information in an easy-to-understand and repeatable manner, but more so due to how her post discussed tool usage.  There any number of DFIR sites out there, web sites and blogs alike, that are full of tool listings...here's this tool, and here's this tool.  Great.  It's like shop class, with all of the tools hanging on the peg board, on their specific hooks, over their specific silhouettes. 

The question folks should have is, great, but how can I best employ those tools to complete the goals of my investigation in a complete, thorough, accurate, and timely manner?

If you're looking for something to blog about...there you go.  There are potentially hundreds or thousands of blog posts based on just that single theme, that one topic.

Sunday, June 17, 2018

Coding in DFIR

A while back, I wrote a blog post where I discussed writing my own tools, and why I do it.  I have publicly shared the tools I've written, and I was asked to share my thoughts as to why I do it, and how it's "of value" to me, and more importantly, to my workflow.  I generally find it very helpful to have tools that fit into and support the workflow I'm using (i.e., timelines), and I also find it very valuable to be able to address issues with the tools (and others) as a result of the available data that's being parsed.

Something I've been interested in for quite some time is the metadata embedded in various file formats used on Windows systems, and this interest has cracked the shell a bit in more than a few cases, and given me just a peak inside of what may be going on when someone creates a file for malicious purposes.  I've not only been able to pull apart OLE format MS documents (MS Word .doc files), but also the OLE objects embedded inside the newer .docx format files.

One tool I like to use is Didier Stevens' oledump.py, in part because it provides a great deal of functionality, particularly where it will decompress VBA macros.  However, I will also use my own OLE parser, because it pulls some metadata that I tend to find very useful when developing a view into an adversary's activities.  Also, there is the potential for additional data to be extracted from areas that other tools simply do not address.

An example of simply metadata extraction came from the Mia Ash persona that Allison Wickoff (an excellent intel analyst at SecureWorks) unraveled.  I didn't have much at all to do with the amazing work that Allison did...all I did was look at the metadata associated with a document sent to the victims.  However, from that small piece of information came some pretty amazing insights.

Another, much older example of how metadata can be extracted and used comes from the Blair case.  That's all I'll say about that one.

Another file format that I've put some work into understanding and unraveling is the LNK file format, which MS has done a very good job of documenting.

There are a number of FOSS tools available for parsing the binary LNK file format that will display the various fields, but I noticed during some recent testing that there were files submitted to VirusTotal that had apparently "done their job" (i.e., been successfully used to infect systems), but the parsing tools failed at various points in dissecting the binary contents of the file.  Some tools that did all of their parsing and then displayed the various fields failed completely, and those that displayed the fields as they were parsed appeared to fail only at a certain point.  Understanding the file format allowed me to identify where the tools appeared to be failing, and in this way, I'm not simply going back to some who wrote a tool with, "..didn't work."  I know exactly what happened and why, and more importantly, can address it.  As a result of the work I did, I have also seen that some aspects of these file formats can be abused without degrading the base functionality offered via the format.

This is where coding in DFIR comes in...using a hex editor, I was able to see where and why the tools were having issues, and I could also see that there really wasn't anything that any of the tools could do about.  What I mean by that is, when parsing Extra Data Blocks within an LNK file (in particular, the TrackerDataBlock) what would be the "machine ID" or NetBIOS name of the system on which the LNK file was created extends beyond the bounds of the size value for that section.  Yes, the machine ID value is of variable length (per the specification) but it also represents the NetBIOS name of a Windows system, so there's an expectation (however inaccurate) that it won't extend beyond the maximum length of a NetBIOS name.

In some of the test cases I observed (files downloaded from VirusTotal using the "tag: lnk" search term), the machine ID field was several bytes longer than expected, pushing the total length of the TrackerDataBlock out beyond the identified length.  This causes all tools to fail, and begin reading the droids from the wrong location.  In other instances, what would be the machine ID is a string of hex characters that extends beyond the length of the TrackerDataBlock, with no apparent droid fields visible via a hex editor.

These are possibly modifications of some kind, or errors in the tool used to create the LNK files (for an example of such a tool, go here).  Either way, could these be considered 'tool marks' that could be used to track LNK files being deployed across campaigns?

Am I suggesting that everyone, every examiner needs to be proficient at coding and knowledgeable of file formats?  No, not at all.  In this blog post, I'm simply elaborating on a previous response, and illustrating how having an understanding of coding can be helpful.

Other examples of how coding in DFIR is useful include not just being able to automate significant portions of your workflow, but also being able to understand what an attacker may have been trying to achieve.  I've engaged in a number of IR engagements over the years where attackers have left batch files or Visual Basic scripts behind, and for the most part, they were easy to understand, and very useful.  However, every now and then, there would be a relatively un- or under-used command that required some research, or something "new".

Another example is decoding weaponized MSWord documents.  Not long ago, I unraveled a macro embedded in an MSWord document, and having some ability to code not only helped me do the unraveling, but also helped me see what was actually being done.  The macro had multiple layers of obfuscation, including base64 encoding, character encoding (i.e., using the character number from an ASCII table instead of the character itself, and decoding via chr()...), as well as a few other tricks.  At one point, there was even a base64-encoded string embedded within a script that was, itself, base64 encoded.

So, in summary, having some capability to code, at some level, is extremely valuable within the DFIR community...that, or knowing someone.  In fact, having someone (or a few someones) you can go to for assistance is just helpful all around.

Tuesday, May 08, 2018

EDR Obviates Compliance

Okay, you'll have to excuse me for the title of this blog post...I needed something to get your attention, and get you to read this, because it's important. I know that stating that "EDR obviates compliance" is going to cause a swirl of emotions in anyone's mind, and my intention was to get you to read on and hear me out; 280 characters simply is not enough to share this line of reasoning.

GDPR is just around the corner, and article 4 of GDPR defines a "personal data breach as"...

...‘personal data breach’ means a breach of security leading to the accidental or unlawful destruction, loss, alteration, unauthorised disclosure of, or access to, personal data transmitted, stored or otherwise processed...

Something to be aware of is that this definition includes ransomware, as it includes "access to" and "alteration of". 

As security professionals have been saying for quite some time now, breaches of your perimeter are going to happen, it's simply inevitable.  However, per GDPR, it's not a reportable breach until something happens with respect to 'personal data'.  The simple fact is that without adequate/suitable instrumentation and visibility, you simply won't know if personal data has been accessed during a breach.  EDR provides that visibility.

Where Are We Now?
Right now, we're "right of breach".  Most organizations are simply unprepared for breaches to occur, and breach identification comes in the form of third-party external notification.  This notification comes weeks or even months after the initial intrusion occurred, during with time the dearth of instrumentation and "evidence decay" makes it nearly impossible to definitively determine what the adversary may have done, with any degree of certainty. 

The 2018 Nuix Black Report states that 98% of those surveyed stated that they could breach the perimeter of a target organization and exfil data in 15 hrs or less.  Compare that to the 2017 Ponemon Institute report that indicated that the 'dwell time' (the time between a breach actually occurring and it being detected) was 191 days, and it's still in the triple digits.  I hardly seems far, but that's reality in today's terms.

Where we need to be is "left of breach".  We know breaches are going to occur, and we know that clean up is going to be messy, so what we need to do is get ahead of the game and be prepared.  It's not unlike boxing...if you step into the ring with gloves on, you know you're going to get hit, so it behooves you to get your gloves up and move around a bit, lessening the impact and effect of your opponent's punches. In cybersecurity, we move "left of breach" by employing monitored EDR (MDR) so that breaches can be detected and responded to early in the adversary's attack cycle, allowing us to contain and eradicate the adversary before they're able to access "critical value data", of CVD.  CVD can consist of "personal data", or PII; credit card data, or PCI; healthcare data, or PHI; or it can be intellectual property, or IP.  Regardless of what your "CVD" is, the point is that you want to detect, respond to and stop the bad guy before they're able to access it, and the only way you can do that is to instrument your infrastructure for visibility.

Costs
Chris Pogue wrote this blog post, addressing the "cost of breach" topic, almost three years ago.  With the help of our wonderful marketing department, I recently gave this webinar on reducing costs associated with breaches.  Things haven't changed...breaches are becoming more expensive; more expensive to clean up, more expensive in terms of down-time and lapses in productivity, and more expensive in terms of legal fees and fines.

While researching information for the webinar, one of the things I ran across was the fines that would have been associated with GDPR, had GDPR been in effect at the time that the Equifax breach.  In 2016, Equifax reportedly earned $3.145B USD.  High end GDPR fines can be "20M euros or 4% of world-wide operating revenue, whichever is greater", bringing the fine to $125.8M USD.  That's just the GDPR fine; that does not include direct costs associated with the IR response and investigation, subsequent notification, nor any indirect costs.

On 1 May, Alabama became the 50th US state to enact a data breach notification law, SB 318.  As you read over the stipulations regarding reporting detailed in the law, keep in mind that for the 50 US states, everyone is different.  If you store or process the personal data for US citizens, the cost associated with simply determining to which states you actually have to report is enormous!  And that's before you even report...it doesn't include the actual cost of notification, it's just the cost of determining to whom (which states) you need to report!

EDR/MDR
So where does endpoint detection and response (EDR) come in?

I've been doing DFIR work for about two decades now, and most of the work I've done has been after an incident has occurred.  Early on, the work involved a system or two thought to be infected or compromised in some manner, but it was always well after the supposed infection or compromise occurred.  In most cases, questions involving things like, "...what data was taken?" simply could not be answered.  Either the information to answer the question didn't exist due to evidence decay, or it didn't exist because that information was was never actually recorded or logged.

EDR/MDR solutions allow you to detect a breach early in the adversary's attack cycle, as one of the things monitored by most, if not all, EDR tools is the processes executed on systems.  Yes, you get a lot of 'normal' user stuff (allowing you to detect fraud and insider threats), but you also get a detailed look at the adversary's activities.

To see some really good examples of process trees showing what various attacks look like, go to Phil Burdette's Twitter feedThis tweet includes an example of a phish, with the user clicking "Enable Content" on a malicious document, leading to the implant being downloaded via bitsadminThis tweet illustrates a great example of credential theft.  I could go on and on about this, but I think that the point has been made...EDR gives you a level of visibility that you don't get through any other instrumentation.

Where I've used this information to great effect was detecting an adversary's activities within a client's infrastructure several years ago; the tool being used at the time alerted on the use of an archive utility known as "Rar".  The adversary had renamed the executable so that it was no longer 'rar.exe', but adding a password to protect the archive is not something the adversary could change, and it's what was detected in this case.  So, we had the fact that data was marshaled and archived for exfiltration (along with the password), and we further was that after exfiltrating the archives, the adversary deleted the archived data.  However, once we got an image of the system we were able to carve unallocated space and recover a number of the archives, and using the adversary's password, we were able to open the recovered archives and see exactly what data was taken. 

Today, the way to address this behavior is that rather than alerting on the condition, instead, block the process itself.  And the same thing is true with other adversary behaviors; intelligence gives us excellent insight into adversary behaviors, and using these behaviors, we can implement our EDR solution to alert on some things, but if other specific conditions are met, prevent those actions from completing.  We can further isolate the adversary by isolating the system on the network.

In March of this year, the City of Atlanta was hit with Samsam ransomware.  Interestingly enough, Secureworks, a security MSSP located within the corporate limits of the city, has a good bit of documentation as to the methodology used by the adversary that deploys the ransomware; had the city had an EDR/MDR solution in place prior to the incident, they would've detected and been able to respond to the initial breach, before the adversary was able to deploy the ransomware. 

Benefits
Given all this, the benefits of employing an EDR/MDR solution are:

1. The solution becomes part of your budget.  Employing an EDR/MDR solution is something you plan for, and include in your annual budget.  Breaches are rarely planned for.

2.  You now have definitive data illustrating the actions taken by an adversary prior to your containment and eradication procedures. You have a record of the commands they ran, the actions they took, and hence their behaviors.  You know exactly where they are within your infrastructure, and you can track back to how they got there, meaning that you can quickly contain and eradicate them.

3. Due to early detection and response, you now have definite data to support not having to report the breach.  Let's say that you detect someone on your network running the "whoami" command; this is a behavior that's been observed in many adversaries.  Alerting on this behavior will allow you to respond to the adversary during their initial recon phase, before they're able to perform credential theft, privilege escalation, and lateral movement.  All of their activity and behavior is recorded, and it's a short list because you've detected and responded to them very early in their attack cycle.  As 'personal data' was not accessed, you've obviated your need to report, and obviated costs associated with fines, notification, clean-up, etc.

Monday, May 07, 2018

Tools, Books, Lessons Learned

Tools
As part of my Sunday morning reading a bit ago, I was fascinated by an evolving tweet thread...Eric recently tweeted some random threat hunting advice involving shimcache data.  In response, Nick tweeted regarding an analytic approach to using shimcache/appcompatcache data, at scale, along with AmCache data.  Nick also provided a link to the FireEye blog post that describes their AppCompatProcessor tool.

I really like this approach, and I think that it's a fantastic step forward, not only in single system analytics, analyzing one system through the traditional, "dead box" analysis, but also with respect to an enterprise-wide approach to threat hunting and response.  In many ways, threat hunting and response is the next logical step in DFIR work, isn't it?  Taking what you learned from one system and expanding it to many systems just kinda makes sense, and it looks as though what the FireEye folks did was take what they learned from many single system examinations and developed a process, with an accompanying tool, from that information.

From even a single system analysis approach, there are still a number of analysts within the community who either aren't parsing the AppCompatCache data as a matter of course, or they're misinterpreting the time stamp data all together. Back in Dec, 2016, Matt Bromiley wrote an article describing some of the important aspects of the AppCompatCache data ("important" with respect to interpretation of the data).  Also, be sure to check out read thoroughly the section named "AppCompatSecrets" here, as well as Luis's article here.  A correct understanding of the data and it's context is important going forward.

From an enterprise perspective, there are tools that do nothing more than collect/parse this data from systems, but do not provide any analytics to the data itself, in short, simply dumping this massive trove of data on the analyst.  The analytic approach described by the FireEye folks is similar in nature to the Hamming distance, something I learned about while taking Dr. Hamming's seminar at NPS; in short, the closer the temporal execution of two entries (smaller Hamming distance), the stronger the tcorr value.  This can be applied to great effect as a threat hunting technique across the enterprise.  In the case of AppCompatCache data, the time stamp does not indicate temporal execution; rather, it's a combination of a flag value and the position of the entry, in the order that it's stored within the Registry value.  The "temporal" aspect, in this case, refers to the location of entries in the data with respect to each other.

This can also be applied on a single system dead box analysis, when you look at it as a system, and include not just the System hive, but the System hive in the RegBack folder, as well as any that exist in VSCs, as well as within a memory dump.

I'd recently blogged about using several Registry and Registry-like artifacts in a much smaller approach to "analytics", looking at the use and correlation of multiple sources of data on a single system, with my base assumption being that of single system dead box analysis.  I figured, hey, it's a place to start.

Getting a Book Published
I recently had an opportunity to read Scar's blog post entitled, Finding a Publisher For Your Book.  I found it fascinating because what she wrote about is not too dissimilar from my own experiences.

I have not read her book, Windows Forensic Cookbook.  In fact, I wasn't aware of it until recently.  Once I found out about it, I checked the Amazon site and didn't see any reviews of the book; I'm not surprised...after all, this is the DFIR "community"...but I did hope to get a sense of the material within the book.  So, this isn't a book review, but rather commentary in solidarity with her experiences.

Like Scar, I've put a lot of work into writing a manuscript, only to get the proofs back for review and thinking...what?  In one instance, every single instance of "plugin" had been changed to "plug-in" in a chapter. Uh...no.  My response back to the copy editor was simply to state at the being of the chapter, "...change it back...all of it."

Scar's comments about expectations caught my attention, as well, particularly when the publisher wanted a chapter review turned around in 48 hrs.  What?!?  This is one of those things you have to be prepared for when working with publishers...they simply do not understand their suppliers, at all.  A publisher or editor works on getting multiple books out, that's what they do all day.  However, those of us writing books and providing material do other things all day, and write books when we have time.  That's right...we do the stuff during the day that makes us qualified to write books, and then actually get to write the books in our copious "free time", after putting in a full day or week of work.  These crazy deadlines are something I've pushed against, more so since I've developed greater credibility with the publisher by writing more books.  When my tech editor is working full days just like everyone else and gets a 35+ page chapter with a demand to have the review returned within 48 hrs, that's simply an unrealistic expectation that needs to be addressed up front.

I've been writing books for a while now...it didn't occur to me just how long I've been doing it until last week.  I have a book in the process of being published right now...I'm waiting on the proofs to come back for review...and I submitted a prospectus recently for another book, and the reviews of the proposal I sent in have started coming back.  One reviewer referred to WFA 1/e, published in 2005; this means that I actually started working on it late in 2003, or really in 2004.  And that wasn't my first book. All of this is to say that when I see someone write a blog post where they share their experiences in writing a book and getting it published, what I find most interesting about it is that nothing seems to have changed.  As such, a lot of what Scar wrote in her blog post rings very true, even to this day.

Finally, I've mentioned this before but I'll say it again...over the years, I've heard stories about issues folks have had working with publishers; I've had some of my own, but I like to think that I've learned from them.  Some folks don't get beyond the proposal stage for their book, and some folks have gotten to a signed contract, but drop out due to apparently arbitrary changes made by the editorial staff, after the contract was signed by both parties.  What I had proposed to the publisher I have worked with is to create a liaison position, one where I would work directly with authors (singles, groups) to help them navigate the apparent labyrinth of going from a blank sheet of paper to a published book, using what I've learned over the years.  This never really went anywhere, in part due to the turnover with the publisher...once I had found a champion for the idea, that person left the company.  The fact that after attempting to do this three times and not succeeding, and that the publisher hasn't come back to me to pursue it, tells me that they (the publisher) are happy with the status quo.

If you're interested in writing a book, you don't have to be.  Read over Scar's blog post, ask questions, etc., before you make a decision to commit to writing a book.  It isn't easy, but it can be done.  If your main fear against writing a book is that someone else is going to read it and be critical, keep in mind that no one will be as passionate as you about what you write.

Lessons Learned
I was engaged in an exchange with a trusted and respected colleague recently, and he said something to me that really struck a chord...he said that if I wanted to progress in the direction we were discussing, I've got to stop "giving stuff away for free".  He was referring to my blog (I think), and his point was well taken.  If you're like me, and really (REALLY) enjoy the work...the discovery, the learning, solving a problem in what may be a unique manner, etc...then what does something like writing books and blog posts get you?  I'm not sure what it gets others, but it doesn't lead to being able to conduct analysis, that's for sure.  I mean, why should it, right...if I put it in writing what I did or would do, any someone else can replicate it (like a recipe for tollhouse cookies) then why reach out and say, "Hey, Harlan...I could really use your help on this...", or "...can you analyze these images for malicious activity..."?