jdk/src/share/classes/javax/management/remote/rmi/NoCallStackClassLoader.java
author emcmanus
Mon, 27 Oct 2008 14:02:40 +0100
changeset 1510 e747d3193ef2
parent 2 90ce3da70b43
child 1639 a97859015238
permissions -rw-r--r--
6763639: Remove "rawtypes" warnings from JMX code Reviewed-by: dfuchs

/*
 * Copyright 2003-2006 Sun Microsystems, Inc.  All Rights Reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.  Sun designates this
 * particular file as subject to the "Classpath" exception as provided
 * by Sun in the LICENSE file that accompanied this code.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
 * CA 95054 USA or visit www.sun.com if you need additional information or
 * have any questions.
 */

package javax.management.remote.rmi;

import java.security.ProtectionDomain;

/**
    <p>A class loader that only knows how to define a limited number
    of classes, and load a limited number of other classes through
    delegation to another loader.  It is used to get around a problem
    with Serialization, in particular as used by RMI (including
    RMI/IIOP).  The JMX Remote API defines exactly what class loader
    must be used to deserialize arguments on the server, and return
    values on the client.  We communicate this class loader to RMI by
    setting it as the context class loader.  RMI uses the context
    class loader to load classes as it deserializes, which is what we
    want.  However, before consulting the context class loader, it
    looks up the call stack for a class with a non-null class loader,
    and uses that if it finds one.  So, in the standalone version of
    javax.management.remote, if the class you're looking for is known
    to the loader of jmxremote.jar (typically the system class loader)
    then that loader will load it.  This contradicts the class-loading
    semantics required.

    <p>We get around the problem by ensuring that the search up the
    call stack will find a non-null class loader that doesn't load any
    classes of interest, namely this one.  So even though this loader
    is indeed consulted during deserialization, it never finds the
    class being deserialized.  RMI then proceeds to use the context
    class loader, as we require.

    <p>This loader is constructed with the name and byte-code of one
    or more classes that it defines, and a class-loader to which it
    will delegate certain other classes required by that byte-code.
    We construct the byte-code somewhat painstakingly, by compiling
    the Java code directly, converting into a string, copying that
    string into the class that needs this loader, and using the
    stringToBytes method to convert it into the byte array.  We
    compile with -g:none because there's not much point in having
    line-number information and the like in these directly-encoded
    classes.

    <p>The referencedClassNames should contain the names of all
    classes that are referenced by the classes defined by this loader.
    It is not necessary to include standard J2SE classes, however.
    Here, a class is referenced if it is the superclass or a
    superinterface of a defined class, or if it is the type of a
    field, parameter, or return value.  A class is not referenced if
    it only appears in the throws clause of a method or constructor.
    Of course, referencedClassNames should not contain any classes
    that the user might want to deserialize, because the whole point
    of this loader is that it does not find such classes.
*/

class NoCallStackClassLoader extends ClassLoader {
    /** Simplified constructor when this loader only defines one class.  */
    public NoCallStackClassLoader(String className,
                                  byte[] byteCode,
                                  String[] referencedClassNames,
                                  ClassLoader referencedClassLoader,
                                  ProtectionDomain protectionDomain) {
        this(new String[] {className}, new byte[][] {byteCode},
             referencedClassNames, referencedClassLoader, protectionDomain);
    }

    public NoCallStackClassLoader(String[] classNames,
                                  byte[][] byteCodes,
                                  String[] referencedClassNames,
                                  ClassLoader referencedClassLoader,
                                  ProtectionDomain protectionDomain) {
        super(null);

        /* Validation. */
        if (classNames == null || classNames.length == 0
            || byteCodes == null || classNames.length != byteCodes.length
            || referencedClassNames == null || protectionDomain == null)
            throw new IllegalArgumentException();
        for (int i = 0; i < classNames.length; i++) {
            if (classNames[i] == null || byteCodes[i] == null)
                throw new IllegalArgumentException();
        }
        for (int i = 0; i < referencedClassNames.length; i++) {
            if (referencedClassNames[i] == null)
                throw new IllegalArgumentException();
        }

        this.classNames = classNames;
        this.byteCodes = byteCodes;
        this.referencedClassNames = referencedClassNames;
        this.referencedClassLoader = referencedClassLoader;
        this.protectionDomain = protectionDomain;
    }

    /* This method is called at most once per name.  Define the name
     * if it is one of the classes whose byte code we have, or
     * delegate the load if it is one of the referenced classes.
     */
    @Override
    protected Class<?> findClass(String name) throws ClassNotFoundException {
        for (int i = 0; i < classNames.length; i++) {
            if (name.equals(classNames[i])) {
                return defineClass(classNames[i], byteCodes[i], 0,
                                   byteCodes[i].length, protectionDomain);
            }
        }

        /* If the referencedClassLoader is null, it is the bootstrap
         * class loader, and there's no point in delegating to it
         * because it's already our parent class loader.
         */
        if (referencedClassLoader != null) {
            for (int i = 0; i < referencedClassNames.length; i++) {
                if (name.equals(referencedClassNames[i]))
                    return referencedClassLoader.loadClass(name);
            }
        }

        throw new ClassNotFoundException(name);
    }

    private final String[] classNames;
    private final byte[][] byteCodes;
    private final String[] referencedClassNames;
    private final ClassLoader referencedClassLoader;
    private final ProtectionDomain protectionDomain;

    /**
     * <p>Construct a <code>byte[]</code> using the characters of the
     * given <code>String</code>.  Only the low-order byte of each
     * character is used.  This method is useful to reduce the
     * footprint of classes that include big byte arrays (e.g. the
     * byte code of other classes), because a string takes up much
     * less space in a class file than the byte code to initialize a
     * <code>byte[]</code> with the same number of bytes.</p>
     *
     * <p>We use just one byte per character even though characters
     * contain two bytes.  The resultant output length is much the
     * same: using one byte per character is shorter because it has
     * more characters in the optimal 1-127 range but longer because
     * it has more zero bytes (which are frequent, and are encoded as
     * two bytes in classfile UTF-8).  But one byte per character has
     * two key advantages: (1) you can see the string constants, which
     * is reassuring, (2) you don't need to know whether the class
     * file length is odd.</p>
     *
     * <p>This method differs from {@link String#getBytes()} in that
     * it does not use any encoding.  So it is guaranteed that each
     * byte of the result is numerically identical (mod 256) to the
     * corresponding character of the input.
     */
    public static byte[] stringToBytes(String s) {
        final int slen = s.length();
        byte[] bytes = new byte[slen];
        for (int i = 0; i < slen; i++)
            bytes[i] = (byte) s.charAt(i);
        return bytes;
    }
}

/*

You can use the following Emacs function to convert class files into
strings to be used by the stringToBytes method above.  Select the
whole (defun...) with the mouse and type M-x eval-region, or save it
to a file and do M-x load-file.  Then visit the *.class file and do
M-x class-string.

;; class-string.el
;; visit the *.class file with emacs, then invoke this function

(defun class-string ()
  "Construct a Java string whose bytes are the same as the current
buffer.  The resultant string is put in a buffer called *string*,
possibly with a numeric suffix like <2>.  From there it can be
insert-buffer'd into a Java program."
  (interactive)
  (let* ((s (buffer-string))
         (slen (length s))
         (i 0)
         (buf (generate-new-buffer "*string*")))
    (set-buffer buf)
    (insert "\"")
    (while (< i slen)
      (if (> (current-column) 61)
          (insert "\"+\n\""))
      (let ((c (aref s i)))
        (insert (cond
                 ((> c 126) (format "\\%o" c))
                 ((= c ?\") "\\\"")
                 ((= c ?\\) "\\\\")
                 ((< c 33)
                  (let ((nextc (if (< (1+ i) slen)
                                   (aref s (1+ i))
                                 ?\0)))
                    (cond
                     ((and (<= nextc ?7) (>= nextc ?0))
                      (format "\\%03o" c))
                     (t
                      (format "\\%o" c)))))
                 (t c))))
      (setq i (1+ i)))
    (insert "\"")
    (switch-to-buffer buf)))

*/