Android S packages.xml乱码的原因

2022-12-10 Android 阅读量

问题背景

客户项目遇到PackageManagerService扫描出现错误导致不停重启，然后进入recovery的问题。查看log是在开机读取Settings(packages.xml）中某个应用的签名信息时出错，系统构造了一个null的签名信息（数组）。导致之后比较签名数组长度时出现空指针。于是想把手机中的packages.xml做修改后模拟问题出现的原因。但是把这个文件pull出来之后发现里面是乱码，切换编码也没用。Android 11的模拟器没有这个问题，于是查看了这个文件的序列化过程，略作纪录。

Android S xml序列化变更

以frameworks/base/services/core/java/com/android/server/pm/Settings.java中writeLPr方法中序列化过程为例，Android S中的序列化工具创建代码如下：

final FileOutputStream fstr = new FileOutputStream(mSettingsFilename);
final TypedXmlSerializer serializer = Xml.resolveSerializer(fstr);
serializer.startDocument(null, true);
serializer.setFeature("http://xmlpull.org/v1/doc/features.html#indent-output", true);

serializer.startTag(null, "packages");

Android R中代码如下:

FileOutputStream fstr = new FileOutputStream(mSettingsFilename);
BufferedOutputStream str = new BufferedOutputStream(fstr);

//XmlSerializer serializer = XmlUtils.serializerInstance();
XmlSerializer serializer = new FastXmlSerializer();
serializer.setOutput(str, StandardCharsets.UTF_8.name());
serializer.startDocument(null, true);
serializer.setFeature("http://xmlpull.org/v1/doc/features.html#indent-output", true);

serializer.startTag(null, "packages");

主要差别就在Xml Serializer的构建上，FastXmlSerializer序列化的xml文件可以正常查看，那TypedXmlSerializer有什么不同呢？先看下TypedXmlSerializer的注释：

/**
 * Specialization of {@link XmlSerializer} which adds explicit methods to
 * support consistent and efficient conversion of primitive data types.
 *
 * @hide
 */
 public interface TypedXmlSerializer extends XmlSerializer {
     //……
 }

也就是所TypedXmlSerializer在序列化过程中会对原生数据类型进行转换。TypedXmlSerializer是个接口，具体构建的对象类型要看Xml.resolveSerializer的实现：

/**
 * Creates a new {@link XmlSerializer} which is optimized for use inside the
 * system, typically by supporting only a basic set of features.
 * <p>
 * This returned instance may be configured to write using an efficient
 * binary format instead of a human-readable text format, depending on
 * device feature flags.
 * <p>
 * To ensure that both formats are detected and transparently handled
 * correctly, you must shift to using both {@link #resolveSerializer} and
 * {@link #resolvePullParser}.
 *
 * @hide
 */
public static @NonNull TypedXmlSerializer resolveSerializer(@NonNull OutputStream out)
        throws IOException {
    final TypedXmlSerializer xml;
    if (ENABLE_BINARY_DEFAULT) {
        xml = newBinarySerializer();
    } else {
        xml = newFastSerializer();
    }
    xml.setOutput(out, StandardCharsets.UTF_8.name());
    return xml;
}

从上述代码来看根据系统ENABLE_BINARY_DEFAULT的配置，可以选择构建BinarySerializer或是FastSerializer。ENABLE_BINARY_DEFAULT的定义如下：

1 2	public static final boolean ENABLE_BINARY_DEFAULT = SystemProperties .getBoolean("persist.sys.binary_xml", true);

所以在调试的时候可以设置persist.sys.binary_xml的值为false然后重启来切换xml serializer并更新xml的内容。

BinarySerializer的注释如下：

/**
 * Serializer that writes XML documents using a custom binary wire protocol
 * which benchmarking has shown to be 4.3x faster and use 2.4x less disk space
 * than {@code Xml.newFastSerializer()} for a typical {@code packages.xml}.
 * <p>
 * The high-level design of the wire protocol is to directly serialize the event
 * stream, while efficiently and compactly writing strongly-typed primitives
 * delivered through the {@link TypedXmlSerializer} interface.
 * <p>
 * Each serialized event is a single byte where the lower half is a normal
 * {@link XmlPullParser} token and the upper half is an optional data type
 * signal, such as {@link #TYPE_INT}.
 * <p>
 * This serializer has some specific limitations:
 * <ul>
 * <li>Only the UTF-8 encoding is supported.
 * <li>Variable length values, such as {@code byte[]} or {@link String}, are
 * limited to 65,535 bytes in length. Note that {@link String} values are stored
 * as UTF-8 on the wire.
 * <li>Namespaces, prefixes, properties, and options are unsupported.
 * </ul>
 */

可见使用新的serializer的目的和Android资源编译的目的类似，一个是节省存储空间，一个是加快访问速度。

Wire protocol（传输协议）更多的是表达传输格式。因为代码层面的数据（链表、队列、二叉树）都是结构化的，但网络层看到的都是二进制流，所以把结构化的数据序列化为二进制流发送出去，并且对方也能以同样的格式反序列化出来，这就是wire protocol。跟把对象存储在文件里，重启后再从文件读出来有点类似。（https://www.zhihu.com/question/413112626）