views:

1715

answers:

4

Swing's JPasswordField has the getPassword() method that returns a char array. My understanding of this is that the array can be zeroed immediately after use so that you do not have sensitive things hanging around in memory for long. The old way to retrieve the password was to use getText(), which returns a String object, but it has been deprecated.

So, my question is why it is actually being used by Java during the retrieval process using getPassword()??? To be clearer, I was debugging my test app for something else**, I followed the calls and bang... getText() in JPasswordField was called and, of course, a nice String object with my password has been created and now is hanging around the memory.

Try it for yourself:

public class PasswordTest() {
    public static void main(String[] args) {
        JFrame frame = new JFrame();
        frame.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
        JPasswordField passField = new JPasswordField();
        pass.addActionListener(new ActionListener() {
            public ActionPerformed(ActionEvent evt) {
                char[] p = passField.getPassword(); // put breakpoint
                // do something with the array
            }
        });
        frame.add(passField);
        frame.setVisible(true);
        frame.pack();
    }
}

Follow up question: is this 'hidden' use of getText() dangerous in any way? Of course a dedicated attacker WILL get your password if it has compromised the system, I am talking about a less dedicated one ;)

**I came across this while I was looking for a way to actually display some sensitive data on a Swing component without using a String object. Apparently there is no way to do it unless I am willing to rewrite part (all?) of the Swing API.. not gonna happen.

+8  A: 

Actually, here's the Sun implementation of getPassword():

public char[] getPassword() {
    Document doc = getDocument();
    Segment txt = new Segment();
    try {
        doc.getText(0, doc.getLength(), txt); // use the non-String API
    } catch (BadLocationException e) {
        return null;
    }
    char[] retValue = new char[txt.count];
    System.arraycopy(txt.array, txt.offset, retValue, 0, txt.count);
    return retValue;
}

The only getText in there is a call to getText(int offset, int length, Segment txt), which calls getChars(int where, int len, Segment txt), which in turn copies characters directly into the Segment's buffer. There are no Strings being created there.

Then, the Segment's buffer is copied into the return value and zeroed out before the method returns.

In other words: There is no extra copy of the password hanging around anywhere. It's perfectly safe as long as you use it as directed.

Michael Myers
beat me to it!
Beau Martínez
I'm surprised; it took me about 15 minutes to spelunk through the Swing code and assure myself that nothing tricky was happening.
Michael Myers
I checked it as well ...
Bruno Ranschaert
Must try harder.
Tom Hawtin - tackline
+3  A: 

The Swing implementation is too complex to check by hand. You want tests.

public class Pwd {
    public static void main(String[] args) {
        java.awt.EventQueue.invokeLater(new Runnable() {
            public void run() {
                new javax.swing.JFrame("Pwd") {{
                    add(new javax.swing.JPasswordField() {
                        @Override public String getText() {
                            System.err.println("Awoooga!!");
                            return super.getText();
                        }
                        {
                            addActionListener(
                                new java.awt.event.ActionListener() {
                                    public void actionPerformed(
                                        java.awt.event.ActionEvent event
                                    ) {
                                        // Nice.
                                    }
                                }
                            );
                        }
                    });
                    setDefaultCloseOperation(DISPOSE_ON_CLOSE);
                    pack();
                    setVisible(true);
                }};
            }
        });
    }
}

Looks like the command string for the (pointless) action event to me. There will be other way to cause the effect as well.

A vaguely modern VM will move objects in memory anyway, so clearing the char[] does not necessarily work.

Tom Hawtin - tackline
"A vaguely modern VM will move objects in memory anyway, so clearing the char[] does not necessarily work." That's a very good point. +1 for that alone, before reading the rest.
Michael Myers
Why would a moving GC make clearing char[] ineffective? Are you saying that it's possible in the instant after the char[] is written with the password but before it is cleared, the GC runs and copies the array to somewhere else, leaving the original copy of the password as "garbage" to be overwritten by later allocations?
Jason Creighton
@Jason Yes. Although I believe typical implementations block zero memory (at least in "Eden"-like spaces) rather than do it at allocation time. It's also more likely that code cockups will cause actual problems - the action source, did you remember to clear the document, getText may create an extra `char[]` (but you don't know so you can't clear it), document implementation may reallocate buffer, etc., etc.
Tom Hawtin - tackline
+3  A: 

Ok, my bad... All the bells started ringing as soon as I saw the call to getText() without noticing that it was actually introduced by me with the Action listener here's a stacktrace

PasswordTest$1.getText() line: 14   
PasswordTest$1(JTextField).fireActionPerformed() line: not available    
PasswordTest$1(JTextField).postActionEvent() line: not available    
JTextField$NotifyAction.actionPerformed(ActionEvent) line: not available    
SwingUtilities.notifyAction(Action, KeyStroke, KeyEvent, Object, int) line: not available

Here is the code used:

 import java.awt.event.*;

 import javax.swing.*;

 public class PasswordTest {
        public static void main(String[] args) {
            JFrame frame = new JFrame();
            frame.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
            final JPasswordField passField = new JPasswordField() {
                @Override
                public String getText() {
                    System.err.println("Awhooa: " + super.getText()); //breakpoint
                    return null;
                }
            };
            passField.addActionListener(new ActionListener() {
                public void actionPerformed(ActionEvent evt) {
                    char[] p = passField.getPassword();
                    System.out.println(p);
                }
            });
            frame.add(passField);
            frame.setVisible(true);
            frame.pack();
        }
    }

And here is the console output:

Awhooa: secret
secret

And for the actual call to getPassword(), maybe I am missing something, but where is Segment's buffer zeroed? I see an array copy, but not a zeroing. The returned array will be zeroed by myself, but Segment's array is still there...

import java.util.Arrays;

public class ZeroingTest {
    public static void main(String[] args) {
        char[] a = {'a','b','c'};
        char[] b = new char[a.length];
        System.arraycopy(a, 0, b, 0, b.length);
        System.out.println("Before zeroing: " + Arrays.toString(a) + " " + Arrays.toString(b));
        Arrays.fill(a, '\0');
        System.out.println("After zeroing: " + Arrays.toString(a) + " " + Arrays.toString(b));
    }
}

And the output:

Before zeroing: [a, b, c] [a, b, c]
After zeroing: [?, ?, ?] [a, b, c]

(I put question marks there because I cannot past unprintable characters)

-M

Good point about the Segment char[].
Tom Hawtin - tackline
You're right. I was posting too quickly and thought I was seeing the Segment's buffer get zeroed, but it doesn't. Perhaps you should file a bug with Sun.
Michael Myers
A: 

**I came across this while I was looking for a way to actually display some sensitive data on a Swing component without using a String object. Apparently there is no way to do it unless I am willing to rewrite part (all?) of the Swing API.. not gonna happen.

You can tell a JPasswordField to display the characters by calling field.setEchoChar('\0'). This retains the rest of the protection offered by JPasswordField (no Strings, no cut/copy).

Draemon