Fix memory corruption in io.Writer interface

When hooked up to logrus and running multiple log.Info() calls (for example),
the bytearray being passed to Write() is reused in the successive calls.  As
this reference was being passed down through the channel, it resulted in
corrupted data and JSON serialization errors.

We fix that here by always making a copy before sending to the channel.

Author: tris

splunk/v2/writer.go

@@ -50,9 +50,13 @@ func (w *Writer) Write(b []byte) (int, error) {
 		w.errors = make(chan error, bufferSize)
 		go w.listen()
 	})
+	// Make a local copy of the bytearray so it doesn't get overwritten by
+	// the next call to Write()
+	var b2 = make([]byte, len(b))
+	copy(b2, b)
 	// Send the data to the channel
 	w.dataChan <- &message{
-		data:      b,
+		data:      b2,
 		writtenAt: time.Now(),
 	}
 	// We don't know if we've hit any errors yet, so just say we're good