Proguard 源码分析(七)混淆

本章我们讲Proguard非常重要的一个步骤:混淆Obfuscator

混淆的目的很明显,是为了混淆语义。

我们能轻车熟路的找到混淆的源头:

Obfuscator 的execute方法

我们来看一下就行混淆器给我们设置了怎样的访问者:

ClassVisitor memberInfoLinker =
            configuration.useUniqueClassMemberNames ?
                (ClassVisitor)new AllMemberVisitor(new MethodLinker()) :
                (ClassVisitor)new BottomClassFilter(new MethodLinker());

我们直接跟到最终的访问者,也就是MethodLinker

public void visitAnyClass(Clazz clazz)
    {
        // Collect all non-private members in this class hierarchy.
        clazz.hierarchyAccept(true, true, true, false,
            new AllMethodVisitor(
            new MemberAccessFilter(0, ClassConstants.INTERNAL_ACC_PRIVATE | ClassConstants.INTERNAL_ACC_STATIC,
            this)));

        // Clean up for the next class hierarchy.
        memberMap.clear();
    }

它的意思是说将自己作为AllMethodVisitor 所有方法中,MemberAccessFilter可接受的访问权限的访问者,

既然它的目的很明显是为了处理方法,我们直接跟到相关的调用:

public void visitAnyMember(Clazz clazz, Member member)
    {
        String name       = member.getName(clazz);
        String descriptor = member.getDescriptor(clazz);
        if (name.equals(ClassConstants.INTERNAL_METHOD_NAME_CLINIT) ||
            name.equals(ClassConstants.INTERNAL_METHOD_NAME_INIT))
        {
            return;
        }
        String key = name + ' ' + descriptor;
        Member otherMember = (Member)memberMap.get(key);

        if (otherMember == null)
        {
            Member thisLastMember = lastMember(member);
            memberMap.put(key, thisLastMember);
        } else {
            link(member, otherMember);
        }
    }

我们看到刚开始它会先取得方法的名称和签名,对于cinit 和init不做处理，注释便是最好的源码解析:

// Special cases: <clinit> and <init> are always kept unchanged.
        // We can ignore them here.

之后将通过签名来找到享元池中的方法Member otherMember = (Member)memberMap.get(key);

if (otherMember == null)
        {
            // Get the last method in the chain.
            Member thisLastMember = lastMember(member);

            // Store the new class method in the map.
            memberMap.put(key, thisLastMember);
        }
        else
        {
            // Link both members.
            link(member, otherMember);
        }

如果没有这个方法,便加入到享元池中,但是之前调用了lastMember

我们来看一下lastMember 方法:

 public static Member lastMember(Member member)
    {
        Member lastMember = member;
        while (lastMember.getVisitorInfo() != null &&
               lastMember.getVisitorInfo() instanceof Member)
        {
            lastMember = (Member)lastMember.getVisitorInfo();
        }

        return lastMember;
    }

也就是说如果它的visitor信息里面如果存在,则直接返回上一个的方法,这种写法很类似于职责链模式，但我只是猜测,只是有职责链的感觉,我们继续往下看,如果方法已经存在了呢？~

我们回到clazz.hierarchyAccept 的参数:

 public void hierarchyAccept(boolean      visitThisClass, //true
                                boolean      visitSuperClass, //true
                                boolean      visitInterfaces, //true
                                boolean      visitSubclasses,//false
                                ClassVisitor classVisitor);

这里面很可能存在一个方法visitThisClass,visitSuperClass,visitInterfaces都存在的情况

所以link(member, otherMember); 的目的就是为了合并到一个共同的职责链源头

好的我们继续混淆的过程:

NameMarker nameMarker = new NameMarker();
        ClassPoolVisitor classPoolvisitor =
            ClassSpecificationVisitorFactory.createClassPoolVisitor(configuration.keep,
                                                                    nameMarker,
                                                                    nameMarker,
                                                                    false,
                                                                    false,
                                                                    true);

前文好几次提到了ClassSpecificationVisitorFactory.createClassPoolVisitor 这个方法,我们不做深究我们直接看NameMarker

public void visitProgramClass(ProgramClass programClass)
    {
        keepClassName(programClass);

        // Make sure any outer class names are kept as well.
        programClass.attributesAccept(this);
    }

public void keepClassName(Clazz clazz)
    {
        ClassObfuscator.setNewClassName(clazz,
                                        clazz.getName());
    }

 static void setNewClassName(Clazz clazz, String name)
    {
        clazz.setVisitorInfo(name);
    }

可见:keepClassName 的目的就是为了往VisitorInfo 设置name 标签

public void visitAnyAttribute(Clazz clazz, Attribute attribute) {}

说明它对属性并没有任何的访问操作:

接着由于库文件没有必要混淆,所以要将库文件保持签名

 libraryClassPool.classesAccept(nameMarker);
        libraryClassPool.classesAccept(new AllMemberVisitor(nameMarker));

接着:

 // Mark attributes that have to be kept.
        AttributeVisitor attributeUsageMarker =
            new NonEmptyAttributeFilter(
            new AttributeUsageMarker());

        AttributeVisitor optionalAttributeUsageMarker =
            configuration.keepAttributes == null ? null :
                new AttributeNameFilter(new ListParser(new NameParser()).parse(configuration.keepAttributes),
                                        attributeUsageMarker);

        programClassPool.classesAccept(
            new AllAttributeVisitor(true,
            new RequiredAttributeFilter(attributeUsageMarker,
                                        optionalAttributeUsageMarker)));

这是对class的属性做标记,接着我们省略掉无关紧要的代码我们直接看混淆的主要访问者

 programClassPool.classesAccept(
            new ClassObfuscator(programClassPool,
                                classNameFactory,
                                packageNameFactory,
                                configuration.useMixedCaseClassNames,
                                configuration.keepPackageNames,
                                configuration.flattenPackageHierarchy,
                                configuration.repackageClasses,
                                configuration.allowAccessModification));

我们可以跟到对于混淆的名是通过:

String name    = programClass.getName();
        String newName = ClassObfuscator.newClassName(programClass);

来生成的而ClassObfuscator.newClassName

实际上是放回访问者的访问标志,而这个标记值是通过ClassObfuscator的

 public void visitProgramClass(ProgramClass programClass)
    {
        // Does this class still need a new name?
        newClassName = newClassName(programClass);
        if (newClassName == null)
        {
            // Make sure the outer class has a name, if it exists. The name will
            // be stored as the new class name, as a side effect, so we'll be
            // able to use it as a prefix.
            programClass.attributesAccept(this);

            // Figure out a package prefix. The package prefix may actually be
            // the an outer class prefix, if any, or it may be the fixed base
            // package, if classes are to be repackaged.
            String newPackagePrefix = newClassName != null ?
                newClassName + ClassConstants.INTERNAL_INNER_CLASS_SEPARATOR :
                newPackagePrefix(ClassUtil.internalPackagePrefix(programClass.getName()));

            // Come up with a new class name, numeric or ordinary.
            newClassName = newClassName != null && numericClassName ?
                generateUniqueNumericClassName(newPackagePrefix) :
                generateUniqueClassName(newPackagePrefix);

            setNewClassName(programClass, newClassName);
        }
    }

来设置的。而对于采用keep标记的那些值,在调用 newClassName(programClass);会放回keep访问者标记的值就不走if中的语句块,这样就达到了只针对非keep中的参数混淆的效果。我们来看一下它是如何给定名字的吧:

private String generateUniqueClassName(String newPackagePrefix)

 private String generateUniqueClassName(String      newPackagePrefix,
                                           NameFactory classNameFactory)
    {
        // Come up with class names until we get an original one.
        String newClassName;
        String newMixedCaseClassName;
        do
        {
            // Let the factory produce a class name.
            newClassName = newPackagePrefix +
                           classNameFactory.nextName();
           
            newMixedCaseClassName = mixedCaseClassName(newClassName);
        }
        while (classNamesToAvoid.contains(newMixedCaseClassName));

        // Explicitly make sure the name isn't used again if we have a
        // user-specified dictionary and we're not allowed to have mixed case
        // class names -- just to protect against problematic dictionaries.
        if (this.classNameFactory != null &&
            !useMixedCaseClassNames)
        {
            classNamesToAvoid.add(newMixedCaseClassName);
        }

        return newClassName;
    }
    {
        // Find the right name factory for this package.
        NameFactory classNameFactory =
            (NameFactory)packagePrefixClassNameFactoryMap.get(newPackagePrefix);
        if (classNameFactory == null)
        {
            // We haven't seen classes in this package before.
            // Create a new name factory for them.
            classNameFactory = new SimpleNameFactory(useMixedCaseClassNames);
            if (this.classNameFactory != null)
            {
                classNameFactory =
                    new DictionaryNameFactory(this.classNameFactory,
                                              classNameFactory);
            }

            packagePrefixClassNameFactoryMap.put(newPackagePrefix,
                                                 classNameFactory);
        }

        return generateUniqueClassName(newPackagePrefix, classNameFactory);
    }

 可见,是通过调用工厂来生成下一个匹配的名字,而且每一个包对应一个工厂,Proguard中有很多的命名工厂类的实现,我们直接用默认的命名工厂类:SimpleNameFactory

 private String name(int index)
    {
        // Which cache do we need?
        List cachedNames = generateMixedCaseNames ?
            cachedMixedCaseNames :
            cachedLowerCaseNames;

        // Do we have the name in the cache?
        if (index < cachedNames.size())
        {
            return (String)cachedNames.get(index);
        }

private String newName(int index)
    {
        // If we're allowed to generate mixed-case names, we can use twice as
        // many characters.
        int totalCharacterCount = generateMixedCaseNames ?
            2 * CHARACTER_COUNT :
            CHARACTER_COUNT;

int baseIndex = index / totalCharacterCount;
        int offset    = index % totalCharacterCount;

        char newChar = charAt(offset);

        String newName = baseIndex == 0 ?
            new String(new char[] { newChar }) :
            (name(baseIndex-1) + newChar);

        return newName;
    }

        // Create a new name and cache it.
        String name = newName(index);
        cachedNames.add(index, name);

        return name;
    }
我们看到它的命名规则非常简单~~就是采用26个字母顺序使用,然后依次扩展.

接下来我们来看下如何将这些规则应用到字节码库中：      

我们看到访问者:

public void visitProgramClass(ProgramClass programClass)
    {
        // Rename this class.
        programClass.thisClassConstantAccept(this);

        // Rename the class members.
        programClass.fieldsAccept(this);
        programClass.methodsAccept(this);
    }

ClassRenamer 它对常量的访问操作是:

public void visitClassConstant(Clazz clazz, ClassConstant classConstant)
    {
        // Update the Class entry if required.
        String newName = ClassObfuscator.newClassName(clazz);
        if (newName != null)
        {
            // Refer to a new Utf8 entry.
            classConstant.u2nameIndex =
                new ConstantPoolEditor((ProgramClass)clazz).addUtf8Constant(newName);
        }
    }

很容易看出,实际上它是将常量池对应的索引数据替换成它的新数据,

我们看下它对member的操作吧:

 public void visitProgramMember(ProgramClass  programClass,
                                     ProgramMember programMember)
    {
        // Has the class member name changed?
        String name    = programMember.getName(programClass);
        String newName = MemberObfuscator.newMemberName(programMember);
        if (newName != null &&
            !newName.equals(name))
        {
            programMember.u2nameIndex =
                new ConstantPoolEditor(programClass).addUtf8Constant(newName);
        }
    }

还是一样的道理,先对member定义个一个新的名字,然后替换掉常量中的数据。