The question received a good answer by @mhaller. I'd say the so-called-puzzle was pretty easy and by just looking at the available c-tors of String one should be able to find out the how part, a
Walkthrough
C-tor of interest is below, if you are to break-in/crack/look for security vulnerability always look for non-final arbitrary classes. The case here is java.nio.charset.Charset
//String
public String(byte bytes[], int offset, int length, Charset charset) {
if (charset == null)
throw new NullPointerException("charset");
checkBounds(bytes, offset, length);
char[] v = StringCoding.decode(charset, bytes, offset, length);
this.offset = 0;
this.count = v.length;
this.value = v;
}
The c-tor offers supposedly-fast way to convert
byte[]
to String by passing the Charset not the chartset name to avoid the lookup chartsetName->charset.
It also allows passing an arbitrary Charset object to create String. Charset main routing converts the content of
java.nio.ByteBuffer
to
CharBuffer
. The CharBuffer may hold a reference to char[] and it's available via
array()
, also the CharBuffer is fully modifiable.
//StringCoding
static char[] decode(Charset cs, byte[] ba, int off, int len) {
StringDecoder sd = new StringDecoder(cs, cs.name());
byte[] b = Arrays.copyOf(ba, ba.length);
return sd.decode(b, off, len);
}
//StringDecoder
char[] decode(byte[] ba, int off, int len) {
int en = scale(len, cd.maxCharsPerByte());
char[] ca = new char[en];
if (len == 0)
return ca;
cd.reset();
ByteBuffer bb = ByteBuffer.wrap(ba, off, len);
CharBuffer cb = CharBuffer.wrap(ca);
try {
CoderResult cr = cd.decode(bb, cb, true);
if (!cr.isUnderflow())
cr.throwException();
cr = cd.flush(cb);
if (!cr.isUnderflow())
cr.throwException();
} catch (CharacterCodingException x) {
// Substitution is always enabled,
// so this shouldn't happen
throw new Error(x);
}
return safeTrim(ca, cb.position(), cs);
}
In order to prevent altering the char[]
the java developers copy the array much like any other String construction (for instance public String(char value[])
). However there is an exception - if no SecurityManager is installed, the char[] is not copied.
//Trim the given char array to the given length
//
private static char[] safeTrim(char[] ca, int len, Charset cs) {
if (len == ca.length
&& (System.getSecurityManager() == null
|| cs.getClass().getClassLoader0() == null))
return ca;
else
return Arrays.copyOf(ca, len);
}
So if there is no SecurityManager it's absolutely possible to have a modifiable CharBuffer/char[] that's being referenced by a String.
Everything looks fine by now - except the byte[]
is also copied (the bold above). This is
where java developers went lazy and massively wrong.
The copy is necessary to prevent the rogue Charset (example above) to be able alter the source byte[]. However, imagine the case of having around 512KB byte[]
buffer that contains few String. Attempting to create a single small, few charts - new String(buf, position, position+32,charset)
resulting in massive 512KB byte[] copy. If the buffer were 1KB or so, the impact will never be truly noticed. With large buffers, the performance hit is really huge, though. The simple fix would be to copy the relevant part.
...or well the designers of java.nio
thought about by introducing read-only Buffers. Simply calling ByteBuffer.asReadOnlyBuffer()
would have been enough (if the Charset.getClassLoader()!=null)*
Sometimes even the guys working on java.lang
can get it totally wrong.
*Class.getClassLoader() returns null for bootstrap classes, i.e. the ones coming with the JVM itself.