Patent Application
20170270319
The present disclosure relates generally to computer systems and in particular to integrity of software code in such systems.
This section is intended to introduce the reader to various aspects of art, which may be related to various aspects of the present disclosure that are described and/or claimed below. This discussion is believed to be helpful in providing the reader with background information to facilitate a better understanding of the various aspects of the present disclosure. Accordingly, it should be understood that these statements are to be read in this light, and not as admissions of prior art.
It is, for various reasons, often desirable to ensure that processing devices execute software that has not been tampered with. To this end, different techniques can be used to protect a software image against tampering attacks. The most common technique is to compute a signature or a checksum over code segments and then verify the signature or checksums at a later stage. Checksums are generally computed and verified without any secret while the generation of a cryptographic signature requires a private key and verification of the signature the corresponding public key.
An example of checksum-based protection is CRC32 for the Portable Executable (PE) format used in the Windows operating system. A PE header contains a CRC32 field that gives the checksum of the corresponding code section. To bypass the protection successfully, an attacker first modifies the code section and then replaces the original checksum with a new value computed over the modified code section. This type of attack is possible since the attacker does not need any secret to update checksums of modified code sections.
Given the weakness of checksums, cryptographic signatures are a preferred solution. The generation of the signature is performed before the code release and uses a private (and thus secret) key. The associated public key is appended to the code and later used to check the code integrity at installation of the code or at runtime. An attacker can still modify the code, but since a correct signature for the code cannot be generated without the private key, the attack fails.
Many solutions exist for checking the integrity of applications delivered and executed in native code, such as the solutions provided by Arxan (GuardIT™), Metaforic (Metafortress™), etc. Native code is a set of assembler instructions directly executable by the processor. The set of instructions does not change after installation, which means that a program integrity value remains the same before and after installation (i.e. remains constant over time). In this case, the signature can be generated beforehand and delivered with the application package.
On the other hand, applications distributed in the form of interpreted code—such as code written in Java, Android DEX code, etc.—comprise intermediate instructions that must be passed through an interpreter before it is executed. Unlike native code, interpreted code can be modified after installation time for optimization purposes. The code modification is generally very dependent on the target platform and is thus not necessarily predictable. If the code is modified, a signature generated upon the interpreted code cannot be used to check code integrity and authenticity dynamically at runtime.
To distribute and install application software onto the previously mentioned Android operating system a file format called APK—Android Application PacKage—is used. To make an APK file, a program for Android is first compiled to an intermediate language, and then its parts are packaged into a compressed archive file (ZIP format). The archive file contains the entire program code in a single DEX (Dalvik EXecutable code) file, various resources (e.g. image files), and the manifest of the APK file. The archive file comprises two additional files: CERT.SF and CERT.RSA. CERT.SF contains cryptographic hashes of all other archive files; CERT.RSA contains the public key used for signature verification. Only CERT.SF is signed with the RSA private key. The RSA signature for the CERT.SF enables validation of the entire content of the APK file during installation. Indeed, all the files mentioned in the CERT.SF file are indirectly signed because CERT.SF contains their hashes. Altering any file before installation would cause an error because the software would detect that a file digest does not match the hash in the CERT.SF file. Alternatively, modifying a cryptographic hash value inside the CERT.SF file (as in the attack against checksum-based verification already described) would lead to an error during the signature verification.
A DEX file header also contains a global checksum for the contents of the DEX file. At the first execution of the application, the Android system uses an optimizer which modifies a DEX interpreted byte code into an optimized machine-instructions sequence called ODEX (Optimized DEX) just in time before execution. The optimizer also updates the checksum. The ODEX file is then stored in a specific repository within the Android file system for future use. The ODEX file then becomes the reference for the application software and, when it is present, the original DEX file is not used anymore.
At runtime, the system may verify the integrity of the application using the ODEX checksum. However this option is not set by default in the Android operating system and the Dalvik machine, which is used to execute ODEX code, does not always check ODEX checksums, since checksum verification has a non-negligible impact on execution performance and boot time.
Android version 5.0 and higher introduced the Android Runtime (ART) which replaces the Dalvik machine. The application are still deployed in DEX code, but at installation time, the DEX code is compiled to native code using the ahead-of-time compilation (AOT) feature. Similar to prior solutions, the AOT compilation on a DEX file results in a binary Executable Linkable Format (ELF) file (which takes the place of the ODEX in prior solutions). The DEX code of the application is then compiled once and then the ELF code is afterward launched each time the application is executed. As ART runs native code directly (ELF code), it brings faster execution of applications and improves the overall power consumption.
Thus it can be seen that in an Android system, the APK signature is verified only at installation time. Furthermore, an APK, even when not signed by a central authority, can be installed on an Android device if the user allows installation of applications coming from untrusted sources. The application developers then use their own self-signed certificates that are not linked to any trusted authority. In that case tampered applications can be resigned and re-installed by any hacker on the Android device unbeknownst to its owner.
As already mentioned, Android applications use an interpreter portable format (DEX). This portable format can execute on a large set of devices with different architectures and characteristics: ARM, x86, MIPS, Little/Big Endian etc. In order to improve performance, the DEX code is modified at installation time or at the first use of the application to produce the ODEX or the ELF binary that is optimized for the target device. During optimization or OAT compilation, various things can be modified in the code: instructions can be replaced by others, the alignment of instructions may be changed, the byte order can be swapped, and so on.
The optimization and OAT compilation then raise a security issue. While the signature of the DEX file can still be verified using the CERT.SF and CERT.RSA, this is not the case for the ODEX and ELF files since they have been modified and their integrity is no more linked to the original DEX signature. In other words, integrity and authenticity can only be verified at installation, but not at runtime since an attacker is able to modify the ODEX and ELF code and update the checksum in the header accordingly.
The system is thus vulnerable to at least two classes of attacks: the remote attack and the root attack. In the remote attack, a downloaded malicious application elevates its privileges and gains system permissions. The malicious application may then tamper with ODEX and ELF files stored on the cache repository of the internal storage. In the root attack, the attacker obtains an Android device, for example by purloining the device or by accessing the device when the owner is absent without locking the device session. The attacker can retrieve an installed application from the device's internal storage through a USB link, modify the application, and then push the modified application back onto the internal storage. For the latter attack to be successful, the device must be “rooted” (i.e. “root access” is required to take control of the device's Android system).
The trust in Android application integrity can thus be broken during the application's life cycle. It is possible to trust what is installed on an Android system, but not necessarily what is running.
It will be appreciated that it is desired to have a solution that overcomes at least part of the problems related to the integrity and authenticity of interpreted code applications. The present disclosure provides such a solution.
In a first aspect, the disclosure is directed to a device for processing an application. The device comprises an interface configured to receive the application, memory configured to store the application and a signed checksum and a processing unit configured to modify the application to obtain a modified application, send a checksum generated for the modified application to a trusted entity, receive a signed checksum corresponding to the sent checksum from the trusted entity, and store the signed checksum in the memory.
Various embodiments of the first aspect include:
That the application is received with a first checksum and that the processing unit is further configured to use the first checksum to verify the integrity of the application.
That the processing unit is configured to use the signed checksum to verify the integrity of the modified application at runtime of the modified application.
That the application is implemented as an interpreted code (DEX) and the modified application is implemented as an optimized interpreted code (ODEX) or the modified application is compiled as a native code (ELF). It is advantageous that the processing unit is configured to replace a checksum for the interpreted code with the signed checksum in a header for the interpreted code or the optimised interpreted code.
That the device is a smartphone or a tablet.
That the trusted entity is implemented in the device. It is advantageous that the trusted entity is configured to store at least one checksum for the application, to verify that the checksum for the modified application matches a stored checksum for the application, and to use a signing key to sign the checksum for the modified application. It is preferred that the signing key is protected using software protection techniques.
That the trusted entity is a separate device and that the interface is further configured to receive the checksum for the modified application from the processing unit and send the checksum for the modified application to the trusted entity, and to receive the signed checksum from the trusted entity and send the signed checksum to the processing unit.
That the processing unit is configured to send an activation code for the application together with the checksum for the modified application.
That the processing unit is configured to receive the signed checksum together with a signing certificate.
In a second aspect, the disclosure is directed to a method for processing an application. A device receives the application, modifies the application to obtain a modified application, sends a checksum generated for the modified application to a trusted entity, receives a signed checksum corresponding to the sent checksum from the trusted entity, and stores the signed checksum in the memory.
Preferred features of the present disclosure will now be described, by way of non-limiting example, with reference to the accompanying drawings, in which
The application provider 120 stores at least one application APK file 122 that can be downloaded by the device 110. The application provider 120 also comprises a hardware processor 124 configured to generate checksums for different ODEX or ELF files that correspond to the application DEX file. These checksums can be generated by installing the DEX file on different test or reference devices and calculate the checksum from the resulting ODEX or ELF files. The application provider 120 is also configured to send the checksums for the different ODEX or ELF files that correspond to the application DEX file to the trusted entity 130.
The trusted entity 130 can be implemented inside the Android OS or on an independent device. The trusted entity 130 comprises memory for storing ODEX or ELF checksums for an application, an interface for receiving an ODEX or ELF checksum from the Android OS on the device 110, a processing unit for verifying that the received ODEX or ELF checksum for an application matches a stored ODEX or ELF checksum for the application, a private signing key 132 to be used for signing ODEX or ELF checksums and an interface for sending a signed ODEX or ELF checksum to the device 110. In case the trusted entity 130 is implemented inside the Android OS, the private signing key is preferably protected using software protection techniques, such as code obfuscation and white-box cryptography, or through the use of specific hardware such as a key-store or a crypto engine.
When the application is to be executed for the first time, a Source Acquisition module reads the content of the ODEX or ELF file into the memory 112, reads the ODEX or ELF checksum (CS) from the DEX header and transmits it, in step S206, to the trusted entity 130. The ODEX or ELF checksum is preferably sent over a protected connection such as a Secure Authenticated Channel. The Source Acquisition module is included in a native library of the application (part of Android application can be developed using code other than Java such as C/C++ language). The Java Native Interface (JNI) enables JAVA code running in a Dalvik Machine to call native libraries delivered with the application.
In case the application requires an activation code to be entered, the checksum could be sent to the remote trusted entity 130 together with the activation code.
The trusted entity 130 preferably checks, in step S208, that the received ODEX or ELF checksum corresponds to one of the stored ODEX or ELF checksums for the application. If this is the case, in step S210 the trusted entity 130 signs the received ODEX or ELF checksum using the private signing key and returns, in step S212, the signed ODEX or ELF checksum to the device 110. The trusted entity 130 can also send a signing certificate comprising a corresponding public key together with the signed ODEX or ELF checksum.
In step S214, the Source Acquisition module receives and stores the signed ODEX or ELF checksum (and, if available and needed, the signing certificate).
The application or the Android OS, having access to a public key corresponding to the private signing key, can then check the integrity of the ODEX or ELF, in step S216, by calculating a checksum for the ODEX or ELF and comparing it to the signed ODEX or ELF checksum. The integrity of the signing certificate can also be verified through the use of a trusted root certificate installed on the device or through the use of a chain of certificates eventually leading to the trusted root certificate.
At subsequent executions of the application or at any time during the current execution of the application, the integrity of the application may be verified the same way as in step S216, i.e. by calculating a checksum for the ODEX or ELF and compare it to the signed ODEX or ELF checksum. To this end, it is advantageous that the option to check the integrity of the ODEX or ELF is set in the Android operating system.
It is noted that the solution is compatible with currently deployed Android systems since the code necessary for runtime integrity verification is either part of the existing Android OS or part of the delivered APK package for the application.
In the present description, the term ‘checksum’ is intended to cover a value that enables verification of whether or not the data for which it was generated has been modified after generation of the checksum. A checksum may thus for example also be a hash value, a Cyclic Redundancy Check (CRC) value or other kind of digest; it is preferred that it is computationally infeasible to obtain the code from the checksum. In addition, while a single checksum has been used for clarity, a plurality of checksums may be used, wherein a checksum may be generated for a distinct part of the code (wherein the different parts may overlap), and that a plurality of checksums for different parts of the code are used to generate a single, global checksum that is used for the comparison. The signature may be any suitable cryptographic signature such as a Hash-based Message Authentication Code (HMAC) or a signature based on for example RSA, Digital Signature Algorithm (DSA) or Elliptic Curve Digital Signature Algorithm (ECDSA).
It will be appreciated that the present solution can counter remote attacks successfully.
Root attacks can also be countered if the trusted entity checks that the received ODEX or ELF checksum corresponds to a ‘legitimate’ code. This is to verify that the received ODEX checksum is not the checksum of an APK comprising modified code (which may be the case if the attacker modifies the code after download from the application provider). For this reason is it preferable for the application provider 120 to send the possible ODEX or ELF checksums to the trusted entity 130; in a variant, it is the trusted entity 130 that generates the different ODEX or ELF checksums by OAT compiling or optimizing for a given target device the DEX code of the application. It is noted that the number of potential checksums depends on a limited set of device hardware parameters (CPU endianness, CPU Symmetric MultiProcessing (SMP) mode, etc.) and thus the number of parameter combinations is limited. For instance, only the SMP mode differs for DEX optimization between a Nexus 7 and a Samsung galaxy tab P5100.
While the present solution has been described in an Android environment, it can be adapted to other operating systems that modify the code during installation without enabling secure integrity verification of the installed application at runtime.
It will thus be appreciated that the present disclosure provides a solution that can enable runtime integrity of applications on Android devices.
Each feature disclosed in the description and (where appropriate) the claims and drawings may be provided independently or in any appropriate combination. Features described as being implemented in hardware may also be implemented in software, and vice versa. Reference numerals appearing in the claims are by way of illustration only and shall have no limiting effect on the scope of the claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 14306918.5 | Nov 2014 | EP | regional |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2015/077832 | 11/26/2015 | WO | 00 |
