今天来粗略看下Java中的类加载是如何实现的。为啥标题用的是OpenJDK呢？看下官网文档的一段描述，
The VM and Java SE class loading libraries share the responsibility for class loading. The VM performs constant pool resolution, linking and initialization for classes and interfaces. The loading phase is a cooperative effort between the VM and specific class loaders (java.lang.classLoader).
也就是说，类加载机制的实现，需要虚拟机与Java的类库来一同完成。Java的类库主要完成loading阶段的工作。
代理机制
类加载的代理机制相信大家也都很清楚了，这边再引用下官网文档的描述，
When the runtime environment needs to load a new class for an application, it looks for the class in the following locations, in order:
- Bootstrap classes: the runtime classes in rt.jar, internationalization classes in i18n.jar, and others.
- Installed extensions: classes in JAR files in the lib/ext directory of the JRE, and in the system-wide, platform-specific extension directory.
- The class path: classes, including classes in JAR files, on paths specified by the system property java.class.path. If a JAR file on the class path has a manifest with the Class-Path attribute, JAR files specified by the Class-Path attribute will be searched also.
相应的，也就有了三种类加载器，
The Java SE class loader hierarchy searches the bootstrap class loader, the extension class loader and the system class loader in that order. The system class loader is the default application class loader, which runs “main” and loads classes from the classpath.
alright，那下面我们就来看看这三种类加载器分别在哪定义了。直接看个简单的栗子，
public static void main(String[] args) throws Exception {
Main m = new Main();
System.out.println(m.getClass().getClassLoader());
System.out.println(m.getClass().getClassLoader().getParent());
System.out.println(m.getClass().getClassLoader().getParent().getParent());
}
根据文档的描述，Main的类加载器就是system class loader（scl），scl的parent应该是extension class loader（ecl），ecl的parent应该是bootstrap class loader（bcl）。那么我们来看下输出结果，
sun.misc.Launcher$AppClassLoader@e6f7d2
sun.misc.Launcher$ExtClassLoader@19836ed
null
也就是说，Launcher$AppClassLoader就是scl，Launcher$ExtClassLoader就是ecl了，二者都是在sun.misc.Launcher中定义。奇怪的是ecl的parent居然是个null，那bcl跑哪去了呢？由于bcl需要在虚拟机启动的时候去加载Bootstrap Classes，所以bcl只能由虚拟机来实现了。HotSpot的bcl实现在classLoader.cpp。
VM初始化
现在我们对scl，ecl算是知其然了，接下来就是要知其所以然。既然虚拟机初始化的时候会使用scl来加载Main-Class，那我们以此为切入点，坑进HotSpot的源码中看看具体是怎么走到Launcher$AppClassLoader来的。
关于虚拟机的初始化可参考笔者翻译的这篇文章，下面直接分享下我读代码的一些相关的结果。虚拟机初始化时会走到下面的逻辑，代码在jni.cpp中，
// We call ClassLoader.getSystemClassLoader to obtain the system class loader.
loader = Handle(THREAD, SystemDictionary::java_system_loader());
也就是说会调用ClassLoader#getSystemClassLoader来拿到scl。OK，scl找到了，继续下去看看会调用scl的哪个方法，逻辑在systemDictionary.cpp，
// Call public unsynchronized loadClass(String) directly for all class loaders
// for parallelCapable class loaders. JDK >=7, loadClass(String, boolean) will
// acquire a class-name based lock rather than the class loader object lock.
// JDK < 7 already acquire the class loader lock in loadClass(String, boolean),
// so the call to loadClassInternal() was not required.
//
// UnsyncloadClass flag means both call loadClass(String) and do
// not acquire the class loader lock even for class loaders that are
// not parallelCapable. This was a risky transitional
// flag for diagnostic purposes only. It is risky to call
// custom class loaders without synchronization.
// WARNING If a custom class loader does NOT synchronizer findClass, or callers of
// findClass, the UnsyncloadClass flag risks unexpected timing bugs in the field.
// Do NOT assume this will be supported in future releases.
//
// Added MustCallLoadClassInternal in case we discover in the field
// a customer that counts on this call
if (MustCallLoadClassInternal && has_loadClassInternal()) {
JavaCalls::call_special(&result,
class_loader,
spec_klass,
vmSymbols::loadClassInternal_name(),
vmSymbols::string_class_signature(),
string,
CHECK_(nh));
} else {
JavaCalls::call_virtual(&result,
class_loader,
spec_klass,
vmSymbols::loadClass_name(),
vmSymbols::string_class_signature(),
string,
CHECK_(nh));
}
所以就是说会调用ClassLoader.getSystemClassLoader.loadClass(String)来加载Main-Class了。
getSystemClassLoader
接下来让我们坑进ClassLoader的源码来看看scl与ecl是怎么来的。
/**
* Returns the system class loader for delegation. This is the default
* delegation parent for new <tt>ClassLoader</tt> instances, and is
* typically the class loader used to start the application.
*
* <p> This method is first invoked early in the runtime's startup
* sequence, at which point it creates the system class loader and sets it
* as the context class loader of the invoking <tt>Thread</tt>.
*
* <p> The default system class loader is an implementation-dependent
* instance of this class.
*
* <p> If the system property "<tt>java.system.class.loader</tt>" is defined
* when this method is first invoked then the value of that property is
* taken to be the name of a class that will be returned as the system
* class loader. The class is loaded using the default system class loader
* and must define a public constructor that takes a single parameter of
* type <tt>ClassLoader</tt> which is used as the delegation parent. An
* instance is then created using this constructor with the default system
* class loader as the parameter. The resulting class loader is defined
* to be the system class loader.
*
* <p> If a security manager is present, and the invoker's class loader is
* not <tt>null</tt> and the invoker's class loader is not the same as or
* an ancestor of the system class loader, then this method invokes the
* security manager's {@link
* SecurityManager#checkPermission(java.security.Permission)
* <tt>checkPermission</tt>} method with a {@link
* RuntimePermission#RuntimePermission(String)
* <tt>RuntimePermission("getClassLoader")</tt>} permission to verify
* access to the system class loader. If not, a
* <tt>SecurityException</tt> will be thrown. </p>
*
* @return The system <tt>ClassLoader</tt> for delegation, or
* <tt>null</tt> if none
*
* @throws SecurityException
* If a security manager exists and its <tt>checkPermission</tt>
* method doesn't allow access to the system class loader.
*
* @throws IllegalStateException
* If invoked recursively during the construction of the class
* loader specified by the "<tt>java.system.class.loader</tt>"
* property.
*
* @throws Error
* If the system property "<tt>java.system.class.loader</tt>"
* is defined but the named class could not be loaded, the
* provider class does not define the required constructor, or an
* exception is thrown by that constructor when it is invoked. The
* underlying cause of the error can be retrieved via the
* {@link Throwable#getCause()} method.
*
* @revised 1.4
*/
@CallerSensitive
public static ClassLoader getSystemClassLoader() {
initSystemClassLoader();
if (scl == null) {
return null;
}
SecurityManager sm = System.getSecurityManager();
if (sm != null) {
checkClassLoaderPermission(scl, Reflection.getCallerClass());
}
return scl;
}
private static synchronized void initSystemClassLoader() {
if (!sclSet) {
if (scl != null)
throw new IllegalStateException("recursive invocation");
sun.misc.Launcher l = sun.misc.Launcher.getLauncher();
if (l != null) {
Throwable oops = null;
scl = l.getClassLoader();
try {
scl = AccessController.doPrivileged(
new SystemClassLoaderAction(scl));
} catch (PrivilegedActionException pae) {
oops = pae.getCause();
if (oops instanceof InvocationTargetException) {
oops = oops.getCause();
}
}
if (oops != null) {
if (oops instanceof Error) {
throw (Error) oops;
} else {
// wrap the exception
throw new Error(oops);
}
}
}
sclSet = true;
}
}
也就是说sun.misc.Launcher.getClassLoader()就是scl，那么看下sun.misc.Launcher，
public Launcher() {
// Create the extension class loader
ClassLoader extcl;
try {
extcl = ExtClassLoader.getExtClassLoader();
} catch (IOException e) {
throw new InternalError(
"Could not create extension class loader");
}
// Now create the class loader to use to launch the application
try {
loader = AppClassLoader.getAppClassLoader(extcl);
} catch (IOException e) {
throw new InternalError(
"Could not create application class loader");
}
// Also set the context class loader for the primordial thread.
Thread.currentThread().setContextClassLoader(loader);
// Finally, install a security manager if requested
String s = System.getProperty("java.security.manager");
if (s != null) {
SecurityManager sm = null;
if ("".equals(s) || "default".equals(s)) {
sm = new java.lang.SecurityManager();
} else {
try {
sm = (SecurityManager)loader.loadClass(s).newInstance();
} catch (IllegalAccessException e) {
} catch (InstantiationException e) {
} catch (ClassNotFoundException e) {
} catch (ClassCastException e) {
}
}
if (sm != null) {
System.setSecurityManager(sm);
} else {
throw new InternalError(
"Could not create SecurityManager: " + s);
}
}
}
alright，到这里已经很清晰了，Launcher.ExtClassLoader和Launcher.AppClassLoader的初始化都找到了。
loadClass
scl找到了，接下来看看loadClass是怎么工作的。Launcher.AppClassLoader和Launcher.ExtClassLoader都继承了URLClassLoader，loadClass方法没有被重写，所以还是在ClassLoader#loadClass，
/**
* Loads the class with the specified <a href="#name">binary name</a>.
* This method searches for classes in the same manner as the {@link
* #loadClass(String, boolean)} method. It is invoked by the Java virtual
* machine to resolve class references. Invoking this method is equivalent
* to invoking {@link #loadClass(String, boolean) <tt>loadClass(name,
* false)</tt>}. </p>
*
* @param name
* The <a href="#name">binary name</a> of the class
*
* @return The resulting <tt>Class</tt> object
*
* @throws ClassNotFoundException
* If the class was not found
*/
public Class<?> loadClass(String name) throws ClassNotFoundException {
return loadClass(name, false);
}
/**
* Loads the class with the specified <a href="#name">binary name</a>. The
* default implementation of this method searches for classes in the
* following order:
*
* <p><ol>
*
* <li><p> Invoke {@link #findLoadedClass(String)} to check if the class
* has already been loaded. </p></li>
*
* <li><p> Invoke the {@link #loadClass(String) <tt>loadClass</tt>} method
* on the parent class loader. If the parent is <tt>null</tt> the class
* loader built-in to the virtual machine is used, instead. </p></li>
*
* <li><p> Invoke the {@link #findClass(String)} method to find the
* class. </p></li>
*
* </ol>
*
* <p> If the class was found using the above steps, and the
* <tt>resolve</tt> flag is true, this method will then invoke the {@link
* #resolveClass(Class)} method on the resulting <tt>Class</tt> object.
*
* <p> Subclasses of <tt>ClassLoader</tt> are encouraged to override {@link
* #findClass(String)}, rather than this method. </p>
*
* <p> Unless overridden, this method synchronizes on the result of
* {@link #getClassLoadingLock <tt>getClassLoadingLock</tt>} method
* during the entire class loading process.
*
* @param name
* The <a href="#name">binary name</a> of the class
*
* @param resolve
* If <tt>true</tt> then resolve the class
*
* @return The resulting <tt>Class</tt> object
*
* @throws ClassNotFoundException
* If the class could not be found
*/
protected Class<?> loadClass(String name, boolean resolve)
throws ClassNotFoundException
{
synchronized (getClassLoadingLock(name)) {
// First, check if the class has already been loaded
Class c = findLoadedClass(name);
if (c == null) {
long t0 = System.nanoTime();
try {
if (parent != null) {
c = parent.loadClass(name, false);
} else {
// 只有ExtClassLoader的parent才是null
// 所以也就只有ExtClassLoader才能代理给BootstrapClassLoader
c = findBootstrapClassOrNull(name);
}
} catch (ClassNotFoundException e) {
// ClassNotFoundException thrown if class not found
// from the non-null parent class loader
}
if (c == null) {
// If still not found, then invoke findClass in order
// to find the class.
long t1 = System.nanoTime();
c = findClass(name);
// this is the defining class loader; record the stats
sun.misc.PerfCounter.getParentDelegationTime().addTime(t1 - t0);
sun.misc.PerfCounter.getFindClassTime().addElapsedTimeFrom(t1);
sun.misc.PerfCounter.getFindClasses().increment();
}
}
if (resolve) {
resolveClass(c);
}
return c;
}
}
scl加载Main-Class的时候向上委托是找不到的，所以就会自己调用findClass了，逻辑在URLClassLoader#findClass，
protected Class<?> findClass(final String name)
throws ClassNotFoundException
{
try {
return AccessController.doPrivileged(
new PrivilegedExceptionAction<Class>() {
public Class run() throws ClassNotFoundException {
String path = name.replace('.', '/').concat(".class");
Resource res = ucp.getResource(path, false);
if (res != null) {
try {
return defineClass(name, res);
} catch (IOException e) {
throw new ClassNotFoundException(name, e);
}
} else {
throw new ClassNotFoundException(name);
}
}
}, acc);
} catch (java.security.PrivilegedActionException pae) {
throw (ClassNotFoundException) pae.getException();
}
}
ucp就是我们定义的classpath，在调用getAppClassLoader方法时初始化，
public static ClassLoader getAppClassLoader(final ClassLoader extcl)
throws IOException
{
final String s = System.getProperty("java.class.path");
final File[] path = (s == null) ? new File[0] : getClassPath(s);
// Note: on bugid 4256530
// Prior implementations of this doPrivileged() block supplied
// a rather restrictive ACC via a call to the private method
// AppClassLoader.getContext(). This proved overly restrictive
// when loading classes. Specifically it prevent
// accessClassInPackage.sun.* grants from being honored.
//
return AccessController.doPrivileged(
new PrivilegedAction<AppClassLoader>() {
public AppClassLoader run() {
URL[] urls =
(s == null) ? new URL[0] : pathToURLs(path);
return new AppClassLoader(urls, extcl);
}
});
}
/*
* Creates a new AppClassLoader
*/
AppClassLoader(URL[] urls, ClassLoader parent) {
super(urls, parent, factory);
}
public URLClassLoader(URL[] urls, ClassLoader parent,
URLStreamHandlerFactory factory) {
super(parent);
// this is to make the stack depth consistent with 1.1
SecurityManager security = System.getSecurityManager();
if (security != null) {
security.checkCreateClassLoader();
}
ucp = new URLClassPath(urls, factory);
acc = AccessController.getContext();
}
从classpath找到类文件并读取后，调用defineClass方法来完成真正的获得Class对象的操作，
/*
* Defines a Class using the class bytes obtained from the specified
* Resource. The resulting Class must be resolved before it can be
* used.
*/
private Class defineClass(String name, Resource res) throws IOException {
long t0 = System.nanoTime();
int i = name.lastIndexOf('.');
URL url = res.getCodeSourceURL();
if (i != -1) {
String pkgname = name.substring(0, i);
// Check if package already loaded.
Manifest man = res.getManifest();
if (getAndVerifyPackage(pkgname, man, url) == null) {
try {
if (man != null) {
definePackage(pkgname, man, url);
} else {
definePackage(pkgname, null, null, null, null, null, null, null);
}
} catch (IllegalArgumentException iae) {
// parallel-capable class loaders: re-verify in case of a
// race condition
if (getAndVerifyPackage(pkgname, man, url) == null) {
// Should never happen
throw new AssertionError("Cannot find package " +
pkgname);
}
}
}
}
// Now read the class bytes and define the class
java.nio.ByteBuffer bb = res.getByteBuffer();
if (bb != null) {
// Use (direct) ByteBuffer:
CodeSigner[] signers = res.getCodeSigners();
CodeSource cs = new CodeSource(url, signers);
sun.misc.PerfCounter.getReadClassBytesTime().addElapsedTimeFrom(t0);
return defineClass(name, bb, cs);
} else {
byte[] b = res.getBytes();
// must read certificates AFTER reading bytes.
CodeSigner[] signers = res.getCodeSigners();
CodeSource cs = new CodeSource(url, signers);
sun.misc.PerfCounter.getReadClassBytesTime().addElapsedTimeFrom(t0);
return defineClass(name, b, 0, b.length, cs);
}
}
继续跟下去最后是会来到ClassLoader的两个本地方法，
private native Class defineClass1(String name, byte[] b, int off, int len,
ProtectionDomain pd, String source);
private native Class defineClass2(String name, java.nio.ByteBuffer b,
int off, int len, ProtectionDomain pd,
String source);
在继续深挖本地方法之前，我们来先做个总结，之前几个xxxClass方法总是看不清楚，现在算是理清了，
loadClass用于实现类加载的代理机制；findClass用于找到类的二进制表示；defineClass用于将类的二进制表示转化成Class对象，这一步由虚拟机来完成；
还有一个resolveClass，是用来手动进行Linking的，
/**
* Links the specified class. This (misleadingly named) method may be
* used by a class loader to link a class. If the class <tt>c</tt> has
* already been linked, then this method simply returns. Otherwise, the
* class is linked as described in the "Execution" chapter of
* <cite>The Java™ Language Specification</cite>.
* </p>
*
* @param c
* The class to link
*
* @throws NullPointerException
* If <tt>c</tt> is <tt>null</tt>.
*
* @see #defineClass(String, byte[], int, int)
*/
protected final void resolveClass(Class<?> c) {
resolveClass0(c);
}
defineClass
两个defineClass方法的本地实现在ClassLoader.c中（如何找到本地方法实现，可参考这篇文章），而两个本地实现最后都会调用JVM_DefineClassWithSource，
/* Define a class with a source (added in JDK1.5) */
JNIEXPORT jclass JNICALL
JVM_DefineClassWithSource(JNIEnv *env, const char *name, jobject loader,
const jbyte *buf, jsize len, jobject pd,
const char *source);
这个方法定义在jvm.h，这里面的接口是对JNI接口（jni.h）的补充。
/*
* This file contains additional functions exported from the VM.
* These functions are complementary to the standard JNI support.
* There are three parts to this file:
*
* First, this file contains the VM-related functions needed by native
* libraries in the standard Java API. For example, the java.lang.Object
* class needs VM-level functions that wait for and notify monitors.
*
* Second, this file contains the functions and constant definitions
* needed by the byte code verifier and class file format checker.
* These functions allow the verifier and format checker to be written
* in a VM-independent way.
*
* Third, this file contains various I/O and nerwork operations needed
* by the standard Java I/O and network APIs.
*/
具体的实现是在jvm.cpp，
JVM_ENTRY(jclass, JVM_DefineClassWithSource(JNIEnv *env, const char *name, jobject loader, const jbyte *buf, jsize len, jobject pd, const char *source))
JVMWrapper2("JVM_DefineClassWithSource %s", name);
return jvm_define_class_common(env, name, loader, buf, len, pd, source, true, THREAD);
JVM_END
// common code for JVM_DefineClass() and JVM_DefineClassWithSource()
// and JVM_DefineClassWithSourceCond()
static jclass jvm_define_class_common(JNIEnv *env, const char *name,
jobject loader, const jbyte *buf,
jsize len, jobject pd, const char *source,
jboolean verify, TRAPS) {
if (source == NULL) source = "__JVM_DefineClass__";
assert(THREAD->is_Java_thread(), "must be a JavaThread");
JavaThread* jt = (JavaThread*) THREAD;
PerfClassTraceTime vmtimer(ClassLoader::perf_define_appclass_time(),
ClassLoader::perf_define_appclass_selftime(),
ClassLoader::perf_define_appclasses(),
jt->get_thread_stat()->perf_recursion_counts_addr(),
jt->get_thread_stat()->perf_timers_addr(),
PerfClassTraceTime::DEFINE_CLASS);
if (UsePerfData) {
ClassLoader::perf_app_classfile_bytes_read()->inc(len);
}
// Since exceptions can be thrown, class initialization can take place
// if name is NULL no check for class name in .class stream has to be made.
TempNewSymbol class_name = NULL;
if (name != NULL) {
const int str_len = (int)strlen(name);
if (str_len > Symbol::max_length()) {
// It's impossible to create this class; the name cannot fit
// into the constant pool.
THROW_MSG_0(vmSymbols::java_lang_NoClassDefFoundError(), name);
}
class_name = SymbolTable::new_symbol(name, str_len, CHECK_NULL);
}
ResourceMark rm(THREAD);
ClassFileStream st((u1*) buf, len, (char *)source);
Handle class_loader (THREAD, JNIHandles::resolve(loader));
if (UsePerfData) {
is_lock_held_by_thread(class_loader,
ClassLoader::sync_JVMDefineClassLockFreeCounter(),
THREAD);
}
Handle protection_domain (THREAD, JNIHandles::resolve(pd));
klassOop k = SystemDictionary::resolve_from_stream(class_name, class_loader,
protection_domain, &st,
verify != 0,
CHECK_NULL);
if (TraceClassResolution && k != NULL) {
trace_class_resolution(k);
}
return (jclass) JNIHandles::make_local(env, Klass::cast(k)->java_mirror());
}
SystemDictionary::resolve_from_stream，
// Add a klass to the system from a stream (called by jni_DefineClass and
// JVM_DefineClass).
// Note: class_name can be NULL. In that case we do not know the name of
// the class until we have parsed the stream.
klassOop SystemDictionary::resolve_from_stream(Symbol* class_name,
Handle class_loader,
Handle protection_domain,
ClassFileStream* st,
bool verify,
TRAPS) {
// Classloaders that support parallelism, e.g. bootstrap classloader,
// or all classloaders with UnsyncloadClass do not acquire lock here
bool DoObjectLock = true;
if (is_parallelCapable(class_loader)) {
DoObjectLock = false;
}
// Make sure we are synchronized on the class loader before we proceed
Handle lockObject = compute_loader_lock_object(class_loader, THREAD);
check_loader_lock_contention(lockObject, THREAD);
ObjectLocker ol(lockObject, THREAD, DoObjectLock);
TempNewSymbol parsed_name = NULL;
// Parse the stream. Note that we do this even though this klass might
// already be present in the SystemDictionary, otherwise we would not
// throw potential ClassFormatErrors.
//
// Note: "name" is updated.
// Further note: a placeholder will be added for this class when
// super classes are loaded (resolve_super_or_fail). We expect this
// to be called for all classes but java.lang.Object; and we preload
// java.lang.Object through resolve_or_fail, not this path.
instanceKlassHandle k = ClassFileParser(st).parseClassFile(class_name,
class_loader,
protection_domain,
parsed_name,
verify,
THREAD);
const char* pkg = "java/";
if (!HAS_PENDING_EXCEPTION &&
!class_loader.is_null() &&
parsed_name != NULL &&
!strncmp((const char*)parsed_name->bytes(), pkg, strlen(pkg))) {
// It is illegal to define classes in the "java." package from
// JVM_DefineClass or jni_DefineClass unless you're the bootclassloader
ResourceMark rm(THREAD);
char* name = parsed_name->as_C_string();
char* index = strrchr(name, '/');
*index = '\0'; // chop to just the package name
while ((index = strchr(name, '/')) != NULL) {
*index = '.'; // replace '/' with '.' in package name
}
const char* fmt = "Prohibited package name: %s";
size_t len = strlen(fmt) + strlen(name);
char* message = NEW_RESOURCE_ARRAY(char, len);
jio_snprintf(message, len, fmt, name);
Exceptions::_throw_msg(THREAD_AND_LOCATION,
vmSymbols::java_lang_SecurityException(), message);
}
if (!HAS_PENDING_EXCEPTION) {
assert(parsed_name != NULL, "Sanity");
assert(class_name == NULL || class_name == parsed_name, "name mismatch");
// Verification prevents us from creating names with dots in them, this
// asserts that that's the case.
assert(is_internal_format(parsed_name),
"external class name format used internally");
// Add class just loaded
// If a class loader supports parallel classloading handle parallel define requests
// find_or_define_instance_class may return a different instanceKlass
if (is_parallelCapable(class_loader)) {
k = find_or_define_instance_class(class_name, class_loader, k, THREAD);
} else {
define_instance_class(k, THREAD);
}
}
// If parsing the class file or define_instance_class failed, we
// need to remove the placeholder added on our behalf. But we
// must make sure parsed_name is valid first (it won't be if we had
// a format error before the class was parsed far enough to
// find the name).
if (HAS_PENDING_EXCEPTION && parsed_name != NULL) {
unsigned int p_hash = placeholders()->compute_hash(parsed_name,
class_loader);
int p_index = placeholders()->hash_to_index(p_hash);
{
MutexLocker mu(SystemDictionary_lock, THREAD);
placeholders()->find_and_remove(p_index, p_hash, parsed_name, class_loader, THREAD);
SystemDictionary_lock->notify_all();
}
return NULL;
}
// Make sure that we didn't leave a place holder in the
// SystemDictionary; this is only done on success
debug_only( {
if (!HAS_PENDING_EXCEPTION) {
assert(parsed_name != NULL, "parsed_name is still null?");
Symbol* h_name = k->name();
Handle h_loader (THREAD, k->class_loader());
MutexLocker mu(SystemDictionary_lock, THREAD);
klassOop check = find_class(parsed_name, class_loader);
assert(check == k(), "should be present in the dictionary");
klassOop check2 = find_class(h_name, h_loader);
assert(check == check2, "name inconsistancy in SystemDictionary");
}
} );
return k();
}
接下去便是到了ClassFileParser::parseClassFile。今天我们暂时先将流程理清楚，更具体的分析且待下回分解：）
接下来让我们来看下findClass这一步中，相应的类二进制表示是怎么被找到的，了解这一步有助于我们解决类似jar包冲突这样的问题。
findClass
在URLClassLoader#findClass中我们可以看到，是通过ucp.getResource(path, false)来找到相应的类文件的，
/**
* Finds the first Resource on the URL search path which has the specified
* name. Returns null if no Resource could be found.
*
* @param name the name of the Resource
* @param check whether to perform a security check
* @return the Resource, or null if not found
*/
public Resource getResource(String name, boolean check) {
if (DEBUG) {
System.err.println("URLClassPath.getResource(\"" + name + "\")");
}
Loader loader;
// 按索引顺序一个一个查找
for (int i = 0; (loader = getLoader(i)) != null; i++) {
Resource res = loader.getResource(name, check);
if (res != null) {
return res;
}
}
return null;
}
/*
* Returns the Loader at the specified position in the URL search
* path. The URLs are opened and expanded as needed. Returns null
* if the specified index is out of range.
*/
private synchronized Loader getLoader(int index) {
if (closed) {
return null;
}
// Expand URL search path until the request can be satisfied
// or the URL stack is empty.
while (loaders.size() < index + 1) {
// Pop the next URL from the URL stack
URL url;
synchronized (urls) {
if (urls.empty()) {
return null;
} else {
url = urls.pop();
}
}
// Skip this URL if it already has a Loader. (Loader
// may be null in the case where URL has not been opened
// but is referenced by a JAR index.)
String urlNoFragString = URLUtil.urlNoFragString(url);
if (lmap.containsKey(urlNoFragString)) {
continue;
}
// Otherwise, create a new Loader for the URL.
Loader loader;
try {
loader = getLoader(url);
// If the loader defines a local class path then add the
// URLs to the list of URLs to be opened.
URL[] urls = loader.getClassPath();
if (urls != null) {
push(urls);
}
} catch (IOException e) {
// Silently ignore for now...
continue;
}
// Finally, add the Loader to the search path.
loaders.add(loader);
lmap.put(urlNoFragString, loader);
}
return loaders.get(index);
}
urls是一个Stack，用来保存classpath。
Stack<URL> urls = new Stack<URL>();
/**
* Creates a new URLClassPath for the given URLs. The URLs will be
* searched in the order specified for classes and resources. A URL
* ending with a '/' is assumed to refer to a directory. Otherwise,
* the URL is assumed to refer to a JAR file.
*
* @param urls the directory and JAR file URLs to search for classes
* and resources
* @param factory the URLStreamHandlerFactory to use when creating new URLs
*/
public URLClassPath(URL[] urls, URLStreamHandlerFactory factory) {
for (int i = 0; i < urls.length; i++) {
path.add(urls[i]);
}
push(urls);
if (factory != null) {
jarHandler = factory.createURLStreamHandler("jar");
}
}
private void push(URL[] us) {
synchronized (urls) {
// 这里是从末尾开始push
// pop的时候就会反过来
for (int i = us.length - 1; i >= 0; --i) {
urls.push(us[i]);
}
}
}
这里有个问题，为什么用Stack这个数据结构？其实是因为classpath即可以使用目录也可以使用jar包（不同的实现有不同的Loader，JarLoader与FileLoader，见getLoader(final URL url)方法），而jar包的处理有一个特别的地方，就是jar包Manifest的Class-Path属性如果指定了其它jar包，需要将它们也加载（参考官网文档），也就是在getLoader(int index)看到的，
URL[] urls = loader.getClassPath();
if (urls != null) {
push(urls);
}
看下JarLoader的实现，
/*
* Returns the JAR file local class path, or null if none.
*/
URL[] getClassPath() throws IOException {
if (index != null) {
return null;
}
if (metaIndex != null) {
return null;
}
ensureOpen();
parseExtensionsDependencies();
if (SharedSecrets.javaUtilJarAccess().jarFileHasClassPathAttribute(jar)) { // Only get manifest when necessary
Manifest man = jar.getManifest();
if (man != null) {
Attributes attr = man.getMainAttributes();
if (attr != null) {
String value = attr.getValue(Name.CLASS_PATH);
if (value != null) {
return parseClassPath(csu, value);
}
}
}
}
return null;
}
所以采用Stack就可以实现深度优先遍历了：）
所以其实是通过遍历URLClassPath#loaders这个ArrayList找到Loader，再由Loader来加载类文件，而URLClassPath#loaders的建立可以看到是有一个深度优先遍历代表classpath的URL数组的过程。
那么接下来的关键就是要看看classpath字符串是如何转换成URL数组的。关键方法是在Launcher#getClassPath，
private static File[] getClassPath(String cp) {
File[] path;
if (cp != null) {
int count = 0, maxCount = 1;
int pos = 0, lastPos = 0;
// Count the number of separators first
while ((pos = cp.indexOf(File.pathSeparator, lastPos)) != -1) {
maxCount++;
lastPos = pos + 1;
}
path = new File[maxCount];
lastPos = pos = 0;
// Now scan for each path component
while ((pos = cp.indexOf(File.pathSeparator, lastPos)) != -1) {
if (pos - lastPos > 0) {
path[count++] = new File(cp.substring(lastPos, pos));
} else {
// empty path component translates to "."
path[count++] = new File(".");
}
lastPos = pos + 1;
}
// Make sure we include the last path component
if (lastPos < cp.length()) {
path[count++] = new File(cp.substring(lastPos));
} else {
path[count++] = new File(".");
}
// Trim array to correct size
if (count != maxCount) {
File[] tmp = new File[count];
System.arraycopy(path, 0, tmp, 0, count);
path = tmp;
}
} else {
path = new File[0];
}
// DEBUG
//for (int i = 0; i < path.length; i++) {
// System.out.println("path[" + i + "] = " + '"' + path[i] + '"');
//}
return path;
}
可以看到是按classpath字符串的字面顺序来构建数组的，所以结论就是findClass也是按照classpath字符串的字面顺序来进行搜索的。
通配符
classpath还有另外一个问题，就是通配符的使用。既然我们已经知道findClass是按照字面顺序来搜索的，那么接下来我们就来看看，如果使用通配符，最后是会被展开成一个什么样的字面顺序。
通配符的展开是由虚拟机的launcher来完成的，这样在虚拟机运行时才能看到展开后的classpath。具体代码在SetClassPath方法，
static void
SetClassPath(const char *s)
{
char *def;
s = JLI_WildcardExpandClasspath(s);
def = JLI_MemAlloc(strlen(s) + 40);
sprintf(def, "-Djava.class.path=%s", s);
AddOption(def, NULL);
}
JLI_WildcardExpandClasspath方法在wildcard.c中，
const char *
JLI_WildcardExpandClasspath(const char *classpath)
{
char *expanded;
FileList fl;
if (strchr(classpath, '*') == NULL)
return classpath;
fl = FileList_split(classpath, PATH_SEPARATOR);
FileList_expandWildcards(fl);
expanded = FileList_join(fl, PATH_SEPARATOR);
FileList_free(fl);
// 设置环境变量可以观察展开后的类路径
// $ set _JAVA_LAUNCHER_DEBUG = 1
if (getenv("_JAVA_LAUNCHER_DEBUG") != 0)
printf("Expanded wildcards:\n"
" before: \"%s\"\n"
" after : \"%s\"\n",
classpath, expanded);
return expanded;
}
在这个文件头可以看到下面的注释，
/*
*
* Expansion of wildcards is done early, prior to the invocation of a
* program's main method, rather than late, during the class-loading
* process itself. Each element of the input class path containing a
* wildcard is replaced by the (possibly empty) sequence of elements
* generated by enumerating the jar files in the named directory. If
* the directory foo contains a.jar, b.jar, and c.jar,
* e.g., then the class path foo/"*" is expanded into
* foo/a.jar:foo/b.jar:foo/c.jar, and that string would be the value
* of the system property java.class.path.
*
* The order in which the jar files in a directory are enumerated in
* the expanded class path is not specified and may vary from platform
* to platform and even from moment to moment on the same machine. A
* well-constructed application should not depend upon any particular
* order. If a specific order is required then the jar files can be
* enumerated explicitly in the class path.
*
*/
其中已经说得很清楚了，通配符的展开并不是确定的，在不同平台可能会不同，即使在同一机器，不同时刻可能也不同，我们不应该依赖通配符展开后的顺序，如果需要依赖顺序就应该直接写明而不要使用通配符。
OK，现在我们又知其然了，但是为什么会这样？接下来继续跟下去，看看在Linux平台上是怎么实现的。可以看到在Linux上面几个关键的方法，
struct WildcardIterator_
{
DIR *dir;
};
static WildcardIterator
WildcardIterator_for(const char *wildcard)
{
DIR *dir;
int wildlen = strlen(wildcard);
if (wildlen < 2) {
dir = opendir(".");
} else {
char *dirname = JLI_StringDup(wildcard);
dirname[wildlen - 1] = '\0';
dir = opendir(dirname);
JLI_MemFree(dirname);
}
if (dir == NULL)
return NULL;
else {
WildcardIterator it = NEW_(WildcardIterator);
it->dir = dir;
return it;
}
}
static char *
WildcardIterator_next(WildcardIterator it)
{
struct dirent* dirp = readdir(it->dir);
return dirp ? dirp->d_name : NULL;
}
所以结论就是在Linux平台上将会使用readdir这个系统调用（参考man）来读取目录下的文件，那么通配符的展开就要取决于这个系统调用读取文件的顺序了。
那么readdir读取文件的顺序又是怎样的？要了解清楚最好还是能坑进Linux源码看看，所以还是留待下次再说吧：）
参考资料
- http://docs.oracle.com/javase/8/docs/technotes/tools/unix/findingclasses.html
- http://docs.oracle.com/javase/8/docs/technotes/tools/unix/classpath.html
- https://docs.oracle.com/javase/tutorial/ext/basics/load.html
- https://docs.oracle.com/javase/specs/jvms/se7/html/jvms-5.html#jvms-5.3
1万+
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
