Guard against read producing a too-long String

Badly-behaved #read implementations may produce String output that
exceeds the requested length, as theorized in jruby#8391. If
this code blindly attempts to copy all bytes from such a String,
it may cause an ArrayIndexOutOfBoundsError when it writes over the
end of the passed-in buffer.

This change guards against such #read implementations by raising
a hard error; we do not maintain a buffer in this class, so we
can't just hold onto the extra bytes, and we don't want to just
silently discard them. A hard error will let a user of this class
find bad #read implementations more quickly and fix the real
problem.

Fixes bug #1 from jruby#8391.

core/src/main/java/org/jruby/util/IOInputStream.java

@@ -141,8 +141,15 @@ public int read(byte[] b, int off, int len) throws IOException {
         if (readValue.isNil()) return -1;
 
         ByteList str = readValue.convertToString().getByteList();
-        System.arraycopy(str.getUnsafeBytes(), str.getBegin(), b, off, str.getRealSize());
 
-        return str.getRealSize();
+        int readSize = str.realSize();
+
+        if (readSize > len) {
+            throw runtime.newIOError("read call of " + len + " bytes produced a String of length " + readSize);
+        }
+
+        System.arraycopy(str.getUnsafeBytes(), str.getBegin(), b, off, readSize);
+
+        return readSize;
      }
 }