Wednesday, September 16, 2015

Rekalling Mimikatz

I'm not really sure that everybody knows that Rekall memory forensics framework contains a Mimikatz plugin: with this post I want to address this shortcoming, since the plugin has many good features and it can be easily extended.

behind the scenes

The act of rekall-ing Mimikatz started when I met Michael Cohen in Prague (SANS DFIR 2014) and a few months later in Dublin (DFRWS 2015). Despite the fact that I learnt so much by speaking with Michael, he deserves the credits to have pushed this plugin development: he released a first version on April 2015, based on what I did with Volatility (see et voilĂ  le mimikatz offline). So by hangout-ing during the night, we co-authored the actual Rekall mimikatz plugin: it was an awesome dive in Windows memory and Rekall internals, guided by Michael who truly has a talent for explaining complicated things in a simple way.

Before going further credits and thanks must go to the awesome reverse engineering research made by Benjamin Delpy: the plugin is based on the knowledge he shared and currently shares.

If you don’t know how (and why) Windows keeps in RAM system and users passwords or hashes, I provide some references: Cached and Stored Credentials Technical Overview; Credentials Processes in Windows Authentication; and, obviously, Gentil Kiwi blog.

debugging symbols power

When Michael showed me the plugin's first version I was amazed since it was really short compared with my Volatility version. Despite the fact my code was a PoC and not an example of well-written Python code for sure, this first version leveraged one of the Rekall key features: the capability to fetch and use in real time Windows debugging symbols.

In order to extract the right information from memory we need both the locations of globals variables and structs layouts. In my Volatility plugin I used byte patterns (provided by Benjamin in his code) to find variables' location in memory: given the fact that Microsoft provides the core symbols we need to achieve our goals (lsasrv!LogonSessionList; lsasrv!hAesKey; lsasrv!h3DesKey; wdigest!l_LogSessList; and so on...), Rekall simplifies  the first half of the job.

This is not the right place to fully address this capability, but once Rekall is instructed to create a profile for a given module (sys, exe or dll), it will extract its GUID from PE debug section and it will fetch from the Windows servers the right symbols by providing the GUID. From this point of view, the Rekall mimikatz plugin behaves like mimilib (the Mimikatz WinDBG extension).

In other words the plugin does not have to search structures in memory using the  byte patterns (or anchors) Benjamin provided: the resulting plugin code is smaller and re-usable, since all supported Windows versions use the same symbols names (more on this later), while anchors based on byte pattern tend to change more frequently. The fact that we don't need to worry about searching for byte patterns makes plugin writing much simpler.

Structs layouts had to be reversed from Mimikatz, without the need to hand code them in the plugin: there is an automated process to pull those from the Mimikatz source code, as explained in the next section.

callable vtypes

Another great capability of the Rekall framework is the vtypes struct definition language, which is not new (see "Rekall Profiles" and "Memory Forensics with Volatility"): vtypes indeed provide a fantastic way to write compact and readable code.  Let me try to show their power with an example in which we used a Mimikatz reversed structure in Rekall.

The widgest module is well-known for keeping in memory the user password: to get the credentials we used Benjamin's _KIWI_WDIGEST_LIST_ENTRY structure defined in kuhl_m_sekurlsa_wdigest.h.

We could have manually written the struct's vtype as it was done in the previous Volatility plugin. However, it is much simpler to automatically extract the vtype definitions from Mimikatz's own debugging information, since Rekall can already parse the PDB file format, it's enough to provide the Mimikatz PDB file to Rekall to get all the structures defined and referenced by the code in it: no manual work! Additionally we don’t have to clutter the source code with inline definitions of these structs - we can simply store the vtype definitions in the Rekall profile repository and fetch them on demand.

We compiled the Mimikatz code with PDB debugging symbols, then we used the parse_pdb Rekall command to get the json files with the structures we needed. Then we pushed the gzipped json file into the profile repository to have it easily available. By doing it for 32 and 64 bits architectures we are able to support both transparently, with really few lines in the plugin! In the next figure the _KIWI_WDIGEST_LIST_ENTRY resulting vtype is shown (64bits).

The previous definition explicitly defines Blink and Flink (backward and forward) pointers for the double linked list, but the Rekall framework has already implemented  re-usable code to parse this type of list: see _LIST_ENTRY class and ListMixIn class defined in "overlays\". Moreover the kiwi structures do not say anything about credentials.  We can extend the vtype by creating an overlay. The overlay "corrects" the generated vtype definition by overriding some fields and adding other fields:

With the previous (and incomplete) overlay, we declare the usage of _LIST_ENTRY (at offset 0) and adding the credential structure. The Cred field contains a _KIWI_GENERIC_PRIMARY_CREDENTIAL struct, located at an offset specified by a callable (which will be evaluated on access to the field) at  8 or 12 bytes (see later) following the end of the LocallyUniqueIdentifier field. With this simple action we are able to get all the list elements with these few lines of code:

That is amazing. The final wdigest code in the plugin becomes incredibly short, and  it's reusable for all Windows versions and architectures known so far!

Note that Rekall profiles are specific to a binary. So each DLL and executable refers to its own profile. It  is a good idea to avoid mixing multiple and possibly different vtype definitions coming from different PDB files. Another point is that we are not using all structures exported by Mimikatz, but only what we really need (see mimikatz_vtypes). Notice that we can simply get the exact address of the global constant "l_LogSessList" directly from debugging symbols, and therefore we do not need to scan for it:

  logons = self.get_constant_object('l_LogSessList', target=...)

before switching to features...

The previous two sections should not be considered a "how to write a Rekall plugin" but an aid to understand the plugin code and why I said it's easy to extend it. Obviously there is a bit more to do to achieve a fully functional plugin, but the Rekall capabilities helps a lot to have a top-down approach, to write less code and to improve readability.


The plugin implements the lsasrv module, which is mandatory to decrypt credentials, the wdigest and livessp modules: the last two SSPs are known to provide the users' passwords if enabled. So we should have good coverage for Windows 7 and 8. Windows 10 will be supported soon. Regarding Windows XP see below.

The plugin logic is the following:

  • switch to lsass process context
  • get crypto material, to be able to decrypt data
  • get all logons to the system and build a LUID dictionary
  • for each LUID get the primary credentials (aka LM, NTLM and SHA1 hashes)
  • if wdigest is used, for each LUID get the credentials 
  • if livessp is used, for each LUID get the credentials
  • for each LUID, get DPAPI master keys from lsasrv
  • render all the data obtained
  • (note: decryption occurs where needed)

To use it just type -f Win7SP1x86.raw mimikatz. If you get some troubles, just add -v parameter to get some details: they are particularly useful when reporting issues. Plugin's results in the following (small) screenshot.

If widgest/livessp is disabled you will unfortunately not get any cleartext passwords from the module. Additional problems include the needed memory being paged out, thus preventing the plugin from achieve its result. You could avoid this problem by using Rekall to dump memory together with the pagefile (use the aff4acquire plugin).

Remember that for DPAPI decryption you only need the SHA1 hash (unless using live logins, for which I will write another post in the future) and that you could find the user's passwords or some hints by accessing lsasecrets and/or decrypting system DPAPI secretes (as WiFi passwords, for example). See my previous post Happy DPAPI.

deprecated xp support

The problem with past XP support was the different type of encryption, DESX, and how the OS was using the DESX key. I spent some time to develop a Python decryption class, which is included in Rekall: for the details see my post UnDesXing. So we have added support for Windows XP/2003! Note that I lacked XP x64 ram images, so in that cases something could go bad: please report and, even better, provide ram images.


I used the Rekall mimikatz plugin in different scenarios and I find it really useful, as much as I enjoyed writing it with Michael. The plugin's main goal is to get users' passwords, or SHA1 hashes to decrypt DPAPI at least. But it could be used from a “pure forensics” point of view, since it lists all logons to the system: actually timestamps are not reported, but it's a matter of adding a "print" statement. Or, when the proper kerberos module will be included, to detect an evil Mimikatz usage.

The plugin is included in Rekall, ready to be used! Just download the latest package from GitHub.

Tuesday, July 28, 2015

Windows Phone PIN cracking

Windows Phone 8 and greater allows the user to lock/unlock the phone by using a numeric PIN code: it's even possible to use a complex alphanumeric password. This post addresses how to obtain the simple numeric PIN code by cracking the authenticator kept in the SOFTWARE hive.

an useless quest?

Actually if you have a physical access to a Windows Phone you don't need the user pincode to examine the user data: with the proper hardware you can usually get a whole dump of the un-encrypted device memory. To my current knowledge the pincode is not used anywhere if not for device locking, so it's almost useless to know it. If the device is under a properly configured MDM, you could face a fully encrypted phone with TPM: in this case you'll have no chance to crack the pincode, even if more testing should be done.

This is exactly what I thought when my colleague Mattia Epifani tried to lure me with the Windows Phone PIN issue: he knows the curious monkey inside me... but I was a reluctant one. He then provided a couple of scenarios where knowing the pincode could be useful: in the end, I traded a couple of beers in return for the pincode cracker.

the starting lead and testbed

Mattia pointed out that the SOFTWARE hive registry key "\Microsoft\Comms\Security\DeviceLock\Object21" was related to the current pincode. This was the start of my travel. I had three physical dumps of three different Windows Phones: two with Windows 8.10 build 341 (label WPB_CXE_R1) and one with Windows 8.00 build 78 (label WP8_CXE_GDR2).

The next figure shows the Object21 registry key content coming from the 1st phone dump.

Indeed the pincode is composed by 5 numbers, but the most attractive value is for sure CredentialHash. In the next figure the bytes blob is reported.

Usually when I face an unknown blob of bytes I tried to figure out, with educated and cautious guesses, if there is a sort of schema: in other words if it represents a structured info. In this case it's quite easy to spot how that blob is organized. The first three DWORDs represent the length of the three following bytes array; the second one is the unicode string SHA256, and the last array has exactly the length of a SHA256 hash. Let me provide a better view of the data.

    80 00 00 00 0E 00 00 00 20 00 00 00 87 A6 A5 93
    5B 2D 8C 55 51 A1 20 07 50 3E A6 48 EB 63 5E CA
    36 9B 4D 5C 65 50 0B 5C 1A 1B E9 34 7B 64 A3 CF
    8B E2 A0 45 5E A0 C3 57 FD 3C 91 AE D8 9F 65 9C
    CE 02 B1 9E 75 06 C7 50 D1 A7 93 ED 76 04 FA 2E
    A4 0A 53 20 1B B1 FD 14 36 C2 2A A9 87 7B C9 BC
    C6 7B 7E 34 A1 EB 2F 6B 33 3A 81 51 99 31 B5 3D
    6F D2 1B 58 69 38 1F 45 5D E3 4B 51 18 36 27 2E
    65 36 3F BB 5B 6A 72 FD F0 D3 38 B7 53 00 48 00
    41 00 32 00 35 00 36 00 00 00 3C DA 9F 6D 42 E8
    83 50 83 4B B2 5E 20 73 7A 4D 66 78 95 01 D0 5A
    5D EA 20 BF 6B B5 53 F6 25 85

a bit of reversing

The question became "who is using that data and how?". Using the simplest approach aka old school effective string searching I got a couple of Windows DLL, SimplePinLap.dll and StrongPinLap.dll, whose names seem self-explanatory. I disclose that I'm really bad with ARM code reversing, but to be short, I spot some truly useful hints inside the SimplePinLap code, by looking for the usual BCryptHashData Win API function.

The 0x80 bytes array is a salt, pseudo-randomly generated. The 0x20 bytes array is the target hash, the one that must match if the inserted pincode is correct. I had some issues in understanding from the assembly how the inserted pin was used in the hashing process, but finally I got the following algorithm (easy, indeed).

HashAlgo(UTF-16-LE-NoTrailing0(pincode) + salt)

a bit of cracking

Finally the pincode seems not stored anywhere inside the system (not considering volatile artifacts): to get user's pincode you need to grab the SOFTWARE hive. If you own such file, you can use my cracking script (link). Just provide it with the hive and all the magic will happen. --software=SOFTWARE

With this script I was able to crack the PIN codes coming out from the devices' dumps I have.

other stuffs

What abouf the StrongPinLap? A blind non-educated guess could state that the same algorithm is used, but the presence of a different library should suggest much more than that. I did not look the strong version: feel free to reach me with a message or a comment, in case you need extra info. Moreover I found some truly interesting stuffs that I will disclose in the next future, so stay tuned.


As I previously said, from my point of view getting the pincode it's an useless or unfeasible task: but I had some fun and I won a couple of beers. Mattia will offer a beer too to the first one who will crack the pincode stored in the blog post example, so hurry up (note: physical meeting needed)!

updates [07.august.2015]

  • Adrian Leong aka @Cheeky4n6Monkey correctly pointed out that the PIN hash data could reside in the Object31 key too. Moreover you could find the CredentialHash value in Object736, Object44... so, it's better to manually inspect all the keys in Microsoft\Comms\Security\DeviceLock. The reasons why the pin hashes can be saved around different keys are currently unknown, and, if you want to crack them, consider the fact that the CredentialActualLength value could be missing. In such cases you'll need to try different pin lengths before getting the result. Adrian provided a different script to crack the pin, which is more flexible but it requires (a bit) manually approach: you can find his script on If you are interested in Windows Phone artifacts check Adrian awesome posts on his blog.
  • Someone asked me if this cracking could be seen as a security vulnerability. No, it's not. You have to physically own the device to grab, when possible, a physical dump or the required files. Different cases, aka device pwned, are out of scope here. In any case a different "vulnerability" must be exploited before pin cracking, which is a quite strong authenticator in this context.
  • Finally if you want to explore Windows Phone from a DFIR point of view, Cindy Murphy, Mattia Epifani and me we'll be speaking on this topic at SANS DFIR Summit in Prague 2015.

Monday, June 22, 2015

A first look at Windows 10 prefetch files

Windows 10 prefetch files (*.pf) show a different file format compared to previous ones.  At first glance you'll spot no textual strings inside, and this was the initial reason that make me try to understand how they changed.

quick&dirty journey

I guess that neither you nor I will run into Windows 10 DFIR cases for a while. That's what I thought when Claudia Meda (@KlodiaMaida) contacted me, showing me a couple of Windows 10 prefetch files. She then provided me some interesting clues that tickled the curious george monkey in me. Officially I do not have spare time, since it's already allocated, so I illegally used the non-existent spare time of spare time: please don't betray me... so I hope you'll tolerate any shortcuts in my quick&dirty journey into the entrails of windows (disgusting, isn't it?).

first lead

First, what a nude prefetch file has to say? Check the first bytes in the next figure, which shows a prefetch file for calc... sorry, now it's calculator (sad, I'll miss you dear calc!). They should remind you some other "prefetch folder" related file: can't you find your memo?

All kidding aside, what the MAM signature recalls is also used by the SuperFetch file format, which on Windows 7 exhibits the very similar MEMO MEM0 signature. A great old (gosh, 2011!) post addressing the SuperFetch file format (and so MEM[0|O] format) is "Windows SuperFetch file format – partial specification" by ReWolf, a worth reading. From there, you can get that the MEM files are, in the first instance, compressed containers and that the Windows API in charge to decompress them is RtlDecompressBuffer.

an actionable lead

I launched windbg inside a Windows 10 virtualized guest (Pro Insider Preview 10.0.10074) and I put a couple of bp on RtlDecompressBuffer and RtlDecompressBufferEx functions: the target process is the SuperFetch process, the svchosted sysmain.dll. To be short, I landed on the moon, which in the case is the sysmain!SmDecompressBuffer method: in the next picture you can see some green boxes (I signed following the decompression branch) and some yellow boxes (checksum check, more on this later).

The routine core is represented in the next picture, where you can (could, if I correctly re-sized the image) easily spot the call to the target method RtlDecompressBufferEx.

When applied to our MAM case, in the end you get three bytes with the magic signature 0x4d4d41, one byte that identifies the compression algorithm used and, eventually, the presence of a checksum: the next 4 bytes are the uncompressed size of the original buffer, then if checksum is in place, you'll get 4 more bytes preceding after [errata 22.06.2015] the uncompressed size that contain the checksum. The remaining data is what must be decompressed with RtlDecompressBufferEx. Which algorithm in used?

The followings are the compression package types and procedures as they are in ntifs.h.

//  Compression package types and procedures.
#define COMPRESSION_FORMAT_NONE          (0x0000)   // winnt
#define COMPRESSION_FORMAT_DEFAULT       (0x0001)   // winnt
#define COMPRESSION_FORMAT_LZNT1         (0x0002)   // winnt
#define COMPRESSION_FORMAT_XPRESS        (0x0003)   // winnt
#define COMPRESSION_FORMAT_XPRESS_HUFF   (0x0004)   // winnt

#define COMPRESSION_ENGINE_STANDARD      (0x0000)   // winnt
#define COMPRESSION_ENGINE_MAXIMUM       (0x0100)   // winnt
#define COMPRESSION_ENGINE_HIBER         (0x0200)   // winnt

So considering that the MAM signature usually is followed by 0x4 (or 0x84), the algorithm is COMPRESSION_FORMAT_XPRESS_HUFF.


To replicate and double checking that findings, I created a small Python Windows native script: with native I pinpoint that you can't use on other OSes different from Windows, since it uses native api calls. Moreover you need Windows 8.1 at least, since the RtlDecompressBufferEx was introduced starting from that OS version. You could use the script to decompress prefetch files, if in need: but you'll get a better solutions at the end of the post. I tweeted about this script some days ago, pointing to a gist I made and that you can find at


In the first instance I ignored the checksum assembly branch, but then I realized that SuperFetch Windows 10 files show the same MAM signature: by applying the previous script, decompression fails. Previously I introduced the fact that a checksum could be present in the prefetch file, when in the third byte (algorithm) you get the most significant bit set: in those cases you see 0x84 as the byte value.

Reconsidering the checksum branch, here is it what happens: the (prefetch|superfetch) file will have 4 more bytes set to the calculated checksum, those bytes stored after the decompresion size (so, bytes 8-11, starting to count from 0): that bytes must be skipped during the decompression phase.

The checksum is a simple CRC32, calculated on the whole file, zeroing out the current file checksum: you can then realize why in the dis-assembly  RtlComputeCrc32 is called three times. I'll updated my Python script to consider that checksum, both on the gist and on the hotoloti github repository.


No, I'm not drunk. yajul (it sounds nice) means Yet Another Joachim Uber Library. Joachim Metz published and currently maintains, among many others, the "Windows Prefetch File (PF) format" document, where he describes the various formats those file use: if you couple it with his "Windows SuperFetch database format", you'll get all the intimate details of Prefetch and SuperFetch files, compression containers included.

Moreover, his libssca (Prefetch files) and libagdb (SuperFetch files) libraries, with the help of libfwnt, are able to correctly handle the decompression and parsing of MAM compression containers (well, the libraries handles all the variants), and that is damned cool!

I want to personally thank Joachim for his prompt support when I reached him with my findings: among other things I got very good suggestions and observations on my short research. I want to share with you an interesting link he provided to me, link to a work made by Jeff Bush on Microsoft Compression Formats.


In the end we get that starting from Windows 10, Prefetch and SuperFetch files are compressed with the XPRESS HUFFMAN algorithm, actually a.k.a. the MAM format. Which is not new: Windows 8.1 uses it to compress SuperFetch files, but not Prefetch files. Moreover from what I see checksum is present only for SuperFetch files and never for Prefetch files.

It remains unclear why Prefetch and SuperFetch files are compressed. Usually compression means space saving (IO reduction?) and computational effort, but it could mean obfuscation too: if you have any clue, I'd be happy to get it.

Anyway with the excellent work made by Joachim we'll be able to understand and to handle those file without any problem. Last but not least, his work his Open Source: not bad, especially in the DFIR world, isn't it?

If you'd need my Python script you can download it from hotoloti or from the gist.

Wednesday, June 3, 2015

iOS 8.3: the end of iOS Forensics?

The latest iOS update (iOS 8.3) is a real nightmare for digital forensics specialists. This article will try to clarify what can you really obtain from an iOS device with iOS 8.3.

As we already know from Jonathan Zdziarski blog, with the introduction of iOS 8 is no longer possible to obtain a so called "Advanced Logical" acquisition based on lockdown service.

However, when we find a device without passocode it is still possible to obtain a backup, although it may be password protected if the user has previously set a password for the local backup.

In the same way we can perform a backup if we find a turned on and locked device, but only if we are able to find a pairing lockdown certificate and the device has been unlocked at least once by the user before the seizure. The same problem about an eventual backup password previously set by the device owner applies to this case too.

The real nightmare is when, and this is the most common case, we have to acquire a device that was turned off.

In this case, also with the lockdown certificate, it is not possible to obtain a backup before unlocking at least once the device. In practical it means that we need to know the passcode to create a backup.

Moreover iOS 8.3 added a new security measure that prevents external tools (not only forensics tools but also iDevice browsing tools like iFunBox) from accessing third party applications sandbox. It means that the only way to read third party contents is to create a backup....but as we already noticed to create a backup you need an unlocked device or a locked device not turned off and a lockdown certificate.

So the question is: what can we really obtain from a locked and turned off device?

It depends, of course, if we are able or not to find a lockdown certificate.

If we haven't got a lockdown certificate we can only recover information about the device and in particular:

  • Device class (iPhone, iPad, iPod)
  • Device name
  • Device color
  • Hardware model
  • iOS version
  • Unique Device ID (UDID)
  • Wi-Fi Address
We can use the tool ideviceinfo that is available in libimobiledvice.

If we have a lockdown certificate there are some more information that we can recover by using the AFC protocol. In particular:

  • Device information
    • IMEI (for devices with telephony capability)
    • Bluetooth Address
    • Language
    • Date and time
    • Timezone
    • Battery charge level
    • Total NAND memory size
    • Empty space size
  • Backup configuration
    • Local vs. iCloud
    • For local backup we can verify if it is password protected
    • Last backup date
  • Installed application list
  • Application distribution on Springboard
  • File contained inside applications that are using iTunes File Sharing. During our test we were able to successfully recover files from:
    • Adobe Reader
    • iZip
    • WinZip
    • File App
    • Office Plus
    • Smart Office
    • USB Disk
  • iBooks folder, containing all the books (both ePub or PDF) saved by the user in the application
  • Downloads folder, containing downloaded files or, in some cases, applications updates
  • iTunes_Control folder
    • iTunes subfolder, containing the iDevice iTunes library list
    • Music subfolder, containing audio files loaded into iTunes library. Files that were originally loaded into iTunes from a PC/MAC are also available
  • Videos loaded into the Device from a PC/MAC
For other folders we can obtain the file name list. For example you can recover it from the DCIM folder and know, at least, if and how many images are stored in the iDevice.

In our test we were able to obtain filename list from:
  • DCIM
  • PhotoStramsData
  • PhotoData
  • Purchases
  • Radio
In the end a short recap:
  • Unlocked device
    • You can always create a local backup
    • If the user has previously set a backup password you need to crack it
  • Locked device
    • Turned on and with lockdown certificate
      • You can create a local backup
      • The same problem as the previous case applies if the user has previously set a backup password
      • Be sure that the device keep charging during the backup process
    • Turned off device and with lockdown certificate
      • Use AFC protocol and recover the most information that you can, as explained in this article
    • Turned on/off device without a lockdown certificate
      • Only device information (name, UDID, etc.)
So, in the end, a suggestion: if you find a turned on device try to isolate it and DO NOT TURN IT OFF before searching for a lockdown certificate on a synced PC/MAC.

For more information on this topic we suggest our book "Learning iOS Forensics", published by PacktPub in March 2015 and authored by Mattia Epifani and Pasquale Stirparo.

Thursday, May 21, 2015


In my own vocabulary, undesxing is the action of decrypting something encrypted with the Microsoft version of the DESX algorithm: a bit obfuscated title but I liked to make a scenographic use of it.

DESX is a variant of the Data Encryption Standard in that a XOR step is added to the plaintext before and after the encryption: you can find a description on wikipedia. So, what is the issue with it? Let me provide the context.

windows lsass

The Windows Local Security Authority (LSA) Subsystem Service (lsass) process is in charge, among other things, to authenticate and log users on to the local system: see Microsoft info here. It's well known that it keeps some sensitive information regarding the logon sessions: for example users' passwords and tokens. This kind of storage - basically due to the SSO capability - is exploited by the never-loved-enough mimikatz, which is able to provide some cool passive (not considering its active operation modalities) information.

This gold mine is kept in lsass process memory, but not all stuffs are in cleartext form. Lsass likes to encrypt some of its (definitely ours!) secrets: obviously lsass has all the data needed to decrypt them. If lsass has that data, we have it too.

Starting from Windows Vista, lsass uses the AES and 3DES to protect that data, without any change to the original algorithms: if you're able to grab the key and the iv (init vector) from the memory (live or dump) you can use any implementation of those algos to decrypt the data.

Before Vista? Lsass processes coming out from the Windows XP family uses different algorithms for the same purpose: RC4 and DESX. Nothing to say about RC4, a couple of caveats about the latter.

some info

Trying to get more technical, when lsass needs to encrypt part of its memory it uses the function LsaEncryptMemory provided by the lsasrv.dll module: the same when it needs to decrypt that data, by simply specifying the modality to use (0 is decrypting, 1 is encrypting). The LsaEncryptMemory has three parameters: the input buffer, the input buffer length and the modality. At this point, both pre-Vista and post-Vista Windows exhibit a similar behaviour: if the input buffer length is not multiple of 8 bytes, RC4 (pre-Vista) and AES  are used; otherwise DESX (pre-Vista) and 3DES.

Let's see what happens inside the Windows XP LsaEncryptMemory in the DESX case: in the following example I used an lsasrv.dll module version 5.1.2600.6058 and GUID FC6285E5-A654-4781-BD4C-BA9E9A0B69E.

There is a call to the CBC function, providing the encryption method as a parameter, desx in figure. The CBC (Cipher Block Chaining mode of operation) method is a bit branched, and I isolated only the relevant path, based on how LsaEncryptMemory initialize its parameters and based on testing.

The "call edx" in the previous figure is a call to the function parameter, desx. In the following the desx decryption branch is reported.

In the next figure a partial view of des decryption branch is reported (truly long, since the loop it's unrolled).

So basically there is a Microsoft interpretation of DES-X algorithm, and that is the first issue to be considered. The second one is the key, which is not present inside the memory. Wait, let me elaborate. During the LsaInitializeProtectedMemory execution the (pseudo)random DES (X) key is created and lost, since the key is exploded in its corresponding rounds-subkeys (key scheduling). So what you will find in memory are the 16 round subkeys, 128 bytes, plus two 8 bytes whitening vectors, for a 144 bytes blob. This is exactly what the debug symbol lsasrv!g_pDESXKey is pointing to, coupled with the lsasrv!g_Feedback symbol (used as the first XOR in CBC function).

LsaEncryptMemory XP transposition ... LsaDecryptXp

Mimikatz reuse the decryption capability offered by the Windows OS: to be able to decrypt DESX encrypted secrets without native Windows features, you need to reverse the algorithm. That's the reason I wrote LsaDecryptXP Python module. I opted for a quick&dirty approach: if you take a look at the code, you'll understand why I used the word transposition. Anyway, a quite good effort/time result. You can find LsaDecryptXP on

who cares?

It depends: one question and one tip. The question: did you known that the awesome Rekall has a mimikatz plugin? The tip: in the next weeks you could check the Rekall code reviews for some cool refactoring of that plugin, and it will use this addon.