Tuesday, December 23, 2014

Reversing APKs for fun and profit: "Decrypting Secret.ly"

I found Secret.ly intersting. this site let's people post-up what's on their minds
without getting exposed, by name or any other details as they ("secret" developers) have been announced.
There is also some bounty for challenging their security related parts of their system, 
which boosted my motivation to try defeating this new way of communication.
So, after stretching my back, and drinking some beer, 2AM, silence at my neighborhood, time to start working :)

Knowledge Requirements:

1.       Little understanding of how does android app is built, should be enough, a bit of Java code understanding.
2.       Little bit knowledge in Reversing.
3.       Some basics in Crypto would help (knowing what is AES should be enough).

In This Post:

1.       Sniffing and decrypting some android HTTPS/SSL stuff.
2.       Reversing/Modding apks (a bit about debugging those also).
3.       Encryption/Decryption in practical usage.
4.       Some cool, pure research and fun J


1. Researching secret.ly mechanism, way of communicating with server-side, 
encryption used, and cookies invloved. understanding of the implementation at whole.
2. Afterwards, managing to challenge the server, trying to:
                a. Find out who's the author of post that is shown to me.
                b. Find security problems with that mechanism, that could make user exposed to attacks, 
                    such as MITM, cookie-stealing and more..

Setup Used:

HTTPS is mainly used here for communicating with the secret api server (api.iheartsecret.com)
therefore, i've arranged:
in my home-pc: 
Installed Fiddler, apktool, smalli, baksmali, jadx, 010Editor (always good), python.
* apktool, baksmali, jadx - used for unpacking/disassembling of apks and dexs, 
which contains the main code and resources for every app, 
jadx is best i've found for converting dex to jars (Java-code)!
* in fiddler, enabled HTTPS decryption of remote client, 
setting proxy port for proxy transportation theough fiddler.
in my android: 
Installed Fiddler certificate as CA, SSH Server, ZipSigner, Apk-Installer.
* in my android i've also configured my wifi connection to use the proxy of fiddler in my home-pc,
so all traffic will go through it.

Why is it enough for sniffing HTTPS (HTTPS-MITM)?

well, how does https work? it verifies the domain certificate, so that the domain you're trying to connect to, as some certificate aythority, that has been signed on it, and that CA (Certificate Authority) is also installed, already in your android, in the section of "Valid Certificates". (There is such in each android OS).
So, when the app is connecting now in https, fiddler will generate certificate for the secret' domain, and because the CA signed on it, is installed manually by us, previously, it will be verified as correct, and the checks will successfully go through.

The Jibbrish flows in...

Let's Start! I've opened up my fiddler, and started sniffing connections to the secret site.
After clicking the app for some notifications, like this:


So i've came-across to some uniquely generated gibbrish, fun-fun-fun:

Well man, that's some unreadable POST-body, ain't it?
Tried check if its:
1.       Gzip – no
2.       Deflate – no
3.       Zlib of some level – no
4.       Encrypted?
Encryption seems the best option, as I've seen this header:

Denote the three parts of it: "ae5" || "34740486…03" || "CW-EnV88...=="
1.       Encryption type: Ae5 ~ "Aes"???
Let's check it up, taking lots of POST-body data (denoted as mi),
And then doing:
gcd(len(m1), len(m2), …, len(mn))
Why you do that? Cause if it’s a block-cipher, as expected, like AES, the cipher data should be aligned to 16bytes size/128bit, which is exactly the block-size of AES.
The result was 16, as expected in bytes, This is AES!
2.       The AES IV? AES nowadays usually uses IV, its some random number, known to all, not a secret, which used to encrypt the message. Here, lets view some random "X-Secret-Enc" headers:
X-Secret-Enc: ae5;3474048679760765703;IhA2j7W7VB1HBHQltjrciQ==
X-Secret-Enc: ae5;3474048679760765703;XquoSMfjKLBlpahxMesciA==
X-Secret-Enc: ae5;3474048679760765703;P7I_R9C-ZKRueWPzpYmqbg==
One can clearly see that only the last part, seems as base64-encoding, is changed in the header, in some session of connection (I've sent some requests for pages in the same secret connection).
After looking over the base64 parts, I've found out that they are really 16bytes of IV data! Doing: base64.urlsafe_decode(last_string), will reveal that information.
Hmm, wise person would ask, why this is the IV and not the KEY? That's a good question my friend, and we cannot be sure that is correct until trying out.
3.       The AES Key? That was the hardest to find, i've assumed it is somehow related to this header, but didn’t really know how, reversing the apk itself was non-effective, but I've managed doing a lot of it, a lot of obfuscation – invalid names, strings encryption, invalid classes, many things probably generated in runtime and more.
But one thing was sure my friend, the second part of the X-Secret-Enc, here it was "3474048679760765703", is the X-Whispr ID, its some runtime generated by server id, found that in the configuration:
<string name="key_sly_user_session_id">3474048679760765703</string>
Notice also, that this part isn’t change each time , made me think it has something related to the key!

Starting Reversing :

This part is little bit more technical, For the advanced reader, so you can actually skip it if you want, it mainly shows my research directions, and various tries, until reaching the correct result. It also shows my usage experience with some programs, and conclusions from working with those.
Got the apk, out of the /data/data/ly.secret… folder, under the name pkg.apk.
So, as, I've said, I used for unpacking all of the above:
The downside
IDA successfully decompiled the classes.dex into
No cross-refs, ida doesn't manage to fully comprehend connection, I've passed that by idapython script to view real xrefs. Also, all data/code is still obfuscated and is dalvik byte-code, and this isn't a really comfortable environment to work in.
·         IDAPython crossrefs can be found easily using:
·    hex(list(XrefsTo(ScreenEA()))[i].frm)
Apktool d <apkname>, with renaming of invalid names
Not all names were remapped.
Decompiled ok to smali files, seems like IDA result, no advantage was in using the latest one here.
Like in IDA.
Unpacked all sources.. seems complete. Nice gui J
Again, ofsucation, couldn't find any part of the string: "X-Secret-Enc" / "Secret-Enc" / "Enc" (in a string)
Works ok, no errors…
didn't unpack some of the sources I've seen using jadx

Example of .java unpacked class: 

This is the SlyRequest, responsible for wrapping all http-requests, done using the HttpOk library.
Its below the HTTPS, and so, all Requests that goes in HTTP to the server side.

Strings searching:
I've searched for many strings, also, in the original files: apk/res/.dex/decrypted resources/unpacked sources/IDA and more…
In all of them, I didn't find any of the words: "X-Secret-Enc" / "Secret-Enc" / "Enc" (in a string) /
I've found although, two strings with 32-bytes length, filled with hex-letters:
            mixpanel key? - '33a52c4d9051b8385d4492ab74fca9c0'
app key? - 'f5d476357695fc86ceed1524cb4218ca'
·         Those aren't worked as well… L
Going over function calls to dec/enc/cipher related:
1.       Found 2 relations to Crypto.Cipher (javax.Cipher and etc.. ) related stuff, but:
a.       Params not found, probably encrypted/obfuscated in some resources. Flow to it wasn't clear.
b.      Usage isn't really clear, why there is SHA256 and also SHA1 / MD5.. also I've seen HMACs flowing there, all references aren't interpreted well in the sources generated, in IDA I can conclude refs, but then going back and forth from sources and IDA is ineffective, and I wasn’t in the mood of writing IDA synchronizing script with the sources file although it's possible.
2.       Done that also for KeySpec class in java and many more…
3.       Finding do_final / init functions, for encryption usuall init/enc callers. Didn’t reach the real key used, again, obfuscation and stuff.
Trying Decrypting with correct-oriented-tries:
Try many ways to decrypt encrypted POST data, none worked:
Denote the three parts of it: "ae5" || "34740486…03" || "CW-EnV88...==" as
Header || SID || IV (after decoding base64)
·         AES(key=md5(SID), iv=IV).decrypt(POST_DATA) ? NO!
·         AES(key=IV, iv=md5(SID)).decrypt(POST_DATA) ? NO!
·         AES(key=key_found_in_ida1, iv=IV).decrypt(POST_DATA) ? NO!
·         AES(key=key_found_in_ida2, iv=IV).decrypt(POST_DATA) ? NO!
·         AES(key=strings_in_IDA, iv=IV).decrypt(POST_DATA) ? NO!
Also, I've tried SHA256 of the SID, with IV given,and even MD5(IV) as the IV. Nothing, all I've got was Gibbrish L

Debugging of the apk?
This was a thought for the future, and wasn't tested, although it can be done with the baksmali/smali, Its called SmaliDebugging, and you can extract and repack your apk in DebuggingMode, cool stuff J
Until this step, I've been really stuck, I've searched all the resources for passwords/keys name/strings related try finding base64/hex strings  of 16/32 bytes, or base64 that can make 16byte keys and more… didn't help.
3 ways to keep going further:
1.       Looking at old Secret.ly apks.
2.       Looking at logs.
3.       Debugging the apk!

Reversing Old Secret:

I had an idea! Why not downloading old-version? Maybe there the app won't be so obfuscated and even relate to the updated current one?
So I've Downloaded old apk of secret, I've found 'X-Secret-Enc' Header! but now with 'ae2' is the start!? whats happening her?!.
Code of encryption there:
  static byte[] encrypt(byte[] bArr, byte[] bArr2) {
            byte[] digest = MessageDigest.getInstance("SHA-256").digest(bArr2);
            byte[] a = Bytes.a(bArr, ZeroBytes); //Bytes.a(a1,a2) = array-appending - a1 to a2. 
                                                                       // ZeroBytes = array of 64bytes of zero.
            byte[] bArr3 = new byte[16];
            Cipher instance = Cipher.getInstance("AES/CBC/PKCS7Padding");
            instance.init(1, new SecretKeySpec(digest, "AES"), new IvParameterSpec(bArr3));
            bArr3 = instance.doFinal(a);
            Key secretKeySpec = new SecretKeySpec(digest, "HmacMD5");
            Mac instance2 = Mac.getInstance(secretKeySpec.getAlgorithm());
            return Bytes.a(bArr3, instance2.doFinal());       }
·         So here, IV={0, … 0} (16 bytes of zero). Hmm.. so probably they just have added iv into the equation.
·         But wait, Key here is new 256bits? weird.. The block size we had was 16-bytes only/128bits?!*
·         There is also HMAC-MD5 appended, with same key used for encryption, over the data.
*I've actually found out that even key in AES is 256bits, the block size always stays 128bits, so the output is always aligned to 128bits/16bytes and not as I thought it should, to 32bytes/256bits if the key is 256bits.

AES Parameters in old APK
    public TypedOutput toBody(Object obj) {
        Log.b("GsonEncryptingConverter", "Magic=" + b[2]); // Logs using android logging-system.
// h = session id string, b[2] = Magic of the current apk J
        return new EncryptedOutput(Crypts.a(bytes, String.format("%d%s", new Object[]{Integer.valueOf(b[2]), h}).getBytes(Charsets.c)), Charsets.c.name());
·         Viola, Key in old version is the sha256 of – "MAGIC || SID" !!!
·         Magic is also printed to logs, in old version, logs can be dumped using ssh/TerminalEmulator app with cmd "logcat –d > logs_dump.txt"
·         But i don't see any logs! not even the "Magic=%s", one.. :(
·        Digging in the code, i've found out, that there is some ke class, that isn't enabling logs on default...
·        Let's Mod that apk up!

"Sweet logs of Mine":

·         looking for debug messages of the program...? to enable logs
·         Logs are controlled in the ke class, using some boolean!
·         Logs are ENABLED ONLY IN "STAGING MODE", not IN OUR "PRODCTION MODE"! so lets change the preference – to true!
·         Added one line – "const v0, 0x1" that’s it! (0x1 = True in dalvik byte-code)
·         REPACKED Using APKTOOL: now I've extra debug info, leaked me some more data on the enc stuff.
o   Just do: apktool d –res apk-path
o   Change smali bytecode in your favorite text editor, add that line to the ke init class.
o   Repack with: apktool b –res folder-to-pack

Logs output Example:

Under-HTTPS Real Encryption Details:

Now, after having the magic, I have tried to assume what the current encryption would like, I WAS CORRECT after some tries, so here is how Secret developers encrypt their POST_DATA J:
sid=741681943761985306 (it just written after the 'ae5' in the X-Secret-Enc block).
key=sha256(magic+sid) # seen that in old secret version with iv=256bytes.
Encryption-Output := AES(KEY=key, IV=iv).encrypt(POST_DATA || zero_array) || HMAC-MD5(KEY=key, data=POST_DATA || zero_array)
·         || means strings-concatenation (e.g:  "A" || "B" == "AB").
·         zero_array is the 64-bytes of zero array.

Examples of Decryption POST-DATA:

>>> AES.new(hashlib.sha256('2006772160' + '741681943761985306').digest(), mode=AES.MODE_CBC, IV=base64.urlsafe_b64decode(b[-1])).decrypt(a[-1][1])
This is standard JSON dictionary, readable, finally! J
*The end of the decryption is trailing of the alignment to 16bytes, done automatically by calling to the AES/CBC/PKCS7Padding.


  Done: Encryption/Decryption/Authentication of packets under HTTPS, in the app-version of the secret.ly site, checked and reversed using my android version 4.4, secret v10.


As i have much spare time as for now, i will continue investigating packets send to/from secret server, hoping to find something intersting, now that i can generate my own malformed packets :)
To Be Continued... :)


·         AES –http://en.wikipedia.org/wiki/Advanced_Encryption_Standard
·         Python libs used – Requests  / Crypto / hashlib / json/ more…
·         APK tools and many more, a real treasure!

Hitcon 2021' CTF - Vulpixelize

Vulpixelize Writeup In this challenge – we are given docker webserver with the following files: 1.       Dockerfile 2.       Simple Flask we...