Java bytecode analyzer customizable via JSON rules. It is a command-line tool that receives a path containing one or more Jar or War files, analyzes them using the provided rules and generates HTML reports with the results.
usage: java -jar cba-cli.jar [OPTIONS] -a DIRECTORY_TO_ANALYZE
-a,--analyze <pathToAnalyze> Path of the directory to run the
analysis.
-c,--checks <checks...> Space separated list of custom checks
that are going to be run in the analysis.
-f,--custom-file <customFile> Specify a file in JSON format to run
custom rules. Read more in
https://github.com/fergarrui/custom-bytecode-analyzer.
-h,--help Print this message.
-i,--items-report <maxItems> Max number of items per report. If the
number of issues found exceeds this
value, the report will be split into
different files. Useful if expecting too
many issues in the report. Default: 50.
-o,--output <outputDir> Directory to save the report. Warning -
if there are already saved reports in
this directory they will be overwritten.
Default is "report".
-v,--verbose-debug Increase verbosity to debug mode.
-vv,--verbose-trace Increase verbosity to trace mode - makes it slower, use it only when you need.
Rules file can be specified using -f,--custom-file argument . The file is in JSON format and has the following structure:
- rules : array(rule)
- name : string
- fields : array(field)
- visibility : (public|protected|private)
- type : string
- valueRegex : string (java regular expression) - only supported if the field is
final - nameRegex : string (java regular expression)
- report : boolean (default: true)
- interfaces : array(string)
- superClass : string
- annotations : array(annotation)
- type : string
- report : boolean (default: true)
- methods : array(method)
- name : string
- visibility : (public|protected|private)
- parameters : array(parameter)
- type : string
- report : boolean (default: true)
- annotations : array(annotation)
- type : string
- report : boolean (default: true)
- variables : array(variable)
- type : string
- nameRegex : string (java regular expression)
- annotations : array(annotation)
- type : string
- report : boolean (default: true)
- report (default: true)
- annotations : array(annotation)
- type : string
- report : boolean (default: true)
- report : boolean (default: true)
- invocations : array(invocation)
- owner : string
- method : method
- name : string
- visibility : (public|protected|private)
- from : method
- name : string
- visibility : (public|protected|private)
- notFrom : method
- name : string
- visibility : (public|protected|private)
- report : boolean (default:true)
You can also check net.nandgr.cba.custom.model.Rules.java to see the structure in Java code.
There are already several rules under the directory examples . Anyway, below are listed examples for every rule.
If we need to find classes with custom deserialization, we can do it quite easily. A class defines custom deserialization by implementing private void readObject(ObjectInputStream in). So we only need to find all classes where that method is defined. It would be enough just to define a rule as:
{
"rules": [{
"name": "Custom deserialization",
"methods": [{
"name": "readObject",
"visibility": "private",
"parameters" : [{
"type" : "java.io.ObjectInputStream"
}]
}]
}]
}It will report methods with private visibility, readObject as name and a parameter of type java.io.ObjectOutputStream. Parameters are an array, if more than one is specified, all of them have to match to be reported. Since we only have one rule, a report named: custom-deserialization-0.html will be created.
In this case, one rule with two methods have to be defined. The same one than in the previous example for deserialization, and a new one to match private void writeObject(ObjectOutputStream out). As shown in the JSON structure above, the property rules.rule.methods is an array of methods, so a rule like this can be written:
{
"rules": [{
"name": "Custom serialization and deserialization",
"methods": [{
"name": "readObject",
"visibility": "private",
"parameters" : [{
"type" : "java.io.ObjectInputStream"
}]
},{
"name": "writeObject",
"report": "false",
"visibility": "private",
"parameters" : [{
"type" : "java.io.ObjectOutputStream"
}]
}]
}]
}The property report was set to false to avoid reporting twice for the same rule. We are using the second method just as a condition, but reporting only readObject methods should be enough for the purpose of this rule.
If a property is not defined, it will always match as true. For example, this rule would return all methods definitions:
{
"rules": [{
"name": "Method definitions",
"methods": [{
}]
}]
}Method invocations can also be found. The JSON in this case would be:
{
"rules": [{
"name": "String equals",
"invocations": [{
"owner": "java.lang.String",
"method": {
"name": "equals"
}
}]
}]
}The property owner specifies the class containing the method.
Another method invocation example a bit more useful than the previous one:
{
"rules": [{
"name": "Method invocation by reflection",
"invocations": [{
"owner": "java.lang.reflect.Method",
"method": {
"name": "invoke"
}
}]
}]
}It is the same than any method invocation, but the name of the method in this case, should be <init>.
{
"rules": [{
"name" : "String instantiation",
"invocations" : [{
"owner" : "java.lang.String",
"method" : {
"name" : "<init>"
}
}]
}]
}This rule will find occurrences of:
[...]
String s = new String("foo");
[...]In this example, we want to find deserialization usages (not classes defining serialization behaviors like in the previous examples). Deserialization happens when ObjectInputStream.readObject() is invoked. for example in this code snippet:
ObjectInputStream in = new ObjectInputStream(fileInputStream);
Object o = in.readObject();So we need to find method invocations from ObjectInputStream named readObject. But it will find a lot of false positives in a researching context, because when a class defines custom deserialization, they make an invocation to this method inside a private void readObject(ObjectInputStream in) method, and that would pollute the report too much. If we want to exclude those cases and keep only genuine deserialization, notFrom property can be used:
{
"rules": [{
"name": "Deserialization usage",
"invocations": [{
"owner": "java.io.ObjectInputStream",
"method": {
"name": "readObject"
},
"notFrom": {
"name": "readObject",
"visibility": "private"
}
}]
}]
}This file will find java.io.ObjectInputStream.readObject() invocations if the invocation is not done inside private void readObject(ObjectInputStream in) method.
A class compiled with this code will not be reported:
private void readObject(ObjectInputStream in) throws IOException, ClassNotFoundException {
Object o = in.readObject();
}But this one will be reported:
public Object deserializeObject(ObjectInputStream in) throws IOException, ClassNotFoundException {
Object o = in.readObject();
return o;
}The property from can be set in invocations in exactly the same way than notFrom, but the result will be the opposite: it will only match if the invocation is made from the defined method.
The property superClass can be used in this case. If we want to find all classes extending javax.servlet.http.HttpServlet, a rule can be:
{
"rules": [{
"name": "Java servlets",
"superClass" : "javax.servlet.http.HttpServlet"
}]
}
A rule can be written to find classes implementing an array of interfaces. if more than one interface is defined in the rule, the class has to implement all of them to be reported. If we want to find classes implementing javax.net.ssl.X509TrustManager, the rule would be:
{
"rules": [{
"name": "X509TrustManager implementations",
"interfaces" : ["javax.net.ssl.X509TrustManager"]
}]
}Please note that interfaces is an array, so make sure you add the strings between square brackets, e.g: ["interface1", "interface2", ...].
Annotations are also supported. Multiple annotations properties can be defined in a rule (finding class annotations), in methods o variables (parameters or local variables). If all of them are found in the analyzed class, it will be reported.
For example, if we want to find Spring endpoints, we would search for classes or methods annotated with org.springframework.web.bind.annotation.RequestMapping. So, the rule can be:
{
"rules": [{
"name": "Spring endpoint - class annotation",
"annotations" : [{
"type" : "org.springframework.web.bind.annotation.RequestMapping"
}]
},
{
"name": "Spring endpoint - method annotation",
"methods" : [{
"annotations" : [{
"type" : "org.springframework.web.bind.annotation.RequestMapping"
}]
}]
}]
}The property rule.fields can be used to find class fields. If we want to find private String fields with password names, a rule like this one could be used:
{
"rules": [{
"name" : "Password fields",
"fields" : [
{
"visibility" : "private",
"type" : "java.lang.String",
"nameRegex" : "(password|pass|psswd|passwd)"
}
]
}]
}To find variables, rule.variables can be used. This property will report local variables and method arguments variables.
If we want to find all variables of type javax.servlet.http.Part, a rule could be:
{
"rules": [{
"name" : "Servlet upload file",
"methods" : [{
"variables" : [{
"type" : "javax.servlet.http.Part"
}]
}]
}]
}Multiple rules can be defined in the same JSON file. They will be processed and reported separately and they will not affect each other. We can combine some of the previous examples rules:
{
"rules": [{
"name": "Custom deserialization",
"methods": [{
"name": "readObject",
"visibility": "private",
"parameters": [{
"type" : "java.io.ObjectInputStream"
}]
}]
},{
"name": "Method invocation by reflection",
"invocations": [{
"owner": "java.lang.reflect.Method",
"method": {
"name": "invoke"
}
}]
}]
}Here, we have two rules ("Custom deserialization" and "Method invocation by reflection"). They will be processed as if you do it in two separated executions. And a report per rule will be generated. If the rules have the same name, they will be reported in the same file.
The project can be downloaded and built to add more complex custom rules in Java code that are not covered by the JSON format. There are already three examples under the package net.nandgr.cba.visitor.checks. Those are CustomDeserializationCheck, DeserializationCheck and InvokeMethodCheck. You can create your own rules by extending net.nandgr.cba.custom.visitor.base.CustomAbstractClassVisitor.
As mentioned above, the reports are created by default under report folder. Every rule will have a separate file unless they have the same name.
If the report is too big, you can split it using the -i,--items-report <maxItems> parameter, each of them will hold the argument specified or less (if it is the last one).
Every reported item, specifies the jar where it is found, the class name and the method name (if it is relevant). It also shows the decompiled version of the class to ease a quick visual check.
Example of how the items are shown for a rule to find java.io.File instantiations:
When searching for security bugs it is very useful to have a call graph. At the moment, a simple DOT compatible file is created under the report directory.
The graph contains all the possible flows where the found issues can be invoked from. For example, if a rule to find deserialization is used,
a graph containing all possible paths leading to the method that calls the deserialization will be generated.
The file is call-graph.dot and it would look like this (this is an extremely simple example):
graph callGraph {
"demo.callgraph.Class1:method1" -- "demo.callgraph.Class2:method2"
"demo.callgraph.Class3:method3" -- "demo.callgraph.Class2:method2"
}
To display it in a visual way, DOT can be used (or any compatible software). For example, to convert the file to svg:
dot -Tsvg call-graph.dot -o call-graph.svg
This is done automatically by default if DOT is found in the system PATH. If not, DOT can be installed in Debian based systems using sudo apt-get install graphviz.
It will create a SVG file named call-graph.svg that can be converted into PNG or visualized using programs like inkscape or just firefox.
A very simple example of the above file call-graph.dot, would be:
There are some limitations, like for example, if the searched item is in a java.lang.Runnable.run() or similar method, it will not find where the thread is executed from.
Also, the graph is cleaning cycles to avoid StackOverflowErrors, it is made in a bit of conservative way so the memory of the system is not drained during an analysis of a large directory.
More options will be added in future versions.
java -jar cba-cli-<version>.jar -a /path/with/jars -f /path/with/json/file/rules.json
To use custom java rules, class names have to be specified as arguments of -c.
java -jar cba-cli-<version>.jar -a /path/with/jars -c DeserializationCheck
Accepts a space separated list, so multiple custom rules can be defined (each of the rules will create a separate report):
java -jar cba-cli-<version>.jar -a /path/with/jars -c DeserializationCheck InvokeMethodCheck CustomDeserializationCheck YourCustomRule
java -jar cba-cli-<version>.jar -a /path/with/jars -f /path/with/json/file/rules.json -c YourCustomRule1 YourCustomRule2
To find errors, verbosity can be increased. Debug level:
java -jar cba-cli-<version>.jar -a /path/with/jars -c YourCustomRule1 -v
Trace level:
java -jar cba-cli-<version>.jar -a /path/with/jars -c YourCustomRule1 -vv
At the moment, the APK has to be converted to JAR first to be analyzed.
- Download dex2jar : https://github.com/pxb1988/dex2jar
- Convert DEX to JAR
d2j-dex2jar.sh -f -o app_to_analyze.jar app_to_analyze.apk
- Run cba-cli.jar as usual passing as
-aparameter the directory containing the converted jar file.
There is already an executable jar file under bin directory at: https://github.com/fergarrui/custom-bytecode-analyzer/blob/master/bin/cba-cli-0.1-SNAPSHOT.jar . If you want to do modifications or add custom rules, the project can be built doing:
git clone https://github.com/fergarrui/custom-bytecode-analyzer.git
cd custom-bytecode-analyzer
mvn clean package
Two jars will be generated under target folder. cba-cli-<version>.jar contains all dependencies and is executable. Can be run using java -jar cba-cli-<version>.jar