[HADOOP-15429] unsynchronized index causes DataInputByteBuffer$Buffer.read hangs

• Type: Bug
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 0.23.0, 2.5.0
• Component/s: io
• Assignee: Unassigned
• Reporter: John Doe
• Created: 30/Apr/18 17:34
• Updated: 02/May/18 01:35

Description

In DataInputByteBuffer$Buffer class, the fields bidx and buffers, etc are unsynchronized when used in read() and reset() function. In certain circumstances, e.g., the reset() is invoked in a loop, the unsynchronized bidx and buffers can trigger a concurrency bug.
Here is the code snippet.

    ByteBuffer[] buffers = new ByteBuffer[0];
    int bidx, pos, length;

    @Override
    public int read(byte[] b, int off, int len) {
      if (bidx >= buffers.length) {
        return -1;
      }
      int cur = 0;
      do {
        int rem = Math.min(len, buffers[bidx].remaining());
        buffers[bidx].get(b, off, rem);
        cur += rem;
        off += rem;
        len -= rem;
      } while (len > 0 && ++bidx < buffers.length); //bidx is unsynchronized
      pos += cur;
      return cur;
    }

    public void reset(ByteBuffer[] buffers) {//if one thread keeps calling reset() in a loop
      bidx = pos = length = 0;
      this.buffers = buffers;
      for (ByteBuffer b : buffers) {
        length += b.remaining();
      }
    }