Stumbling over the Urgent Pointer code in tcp_recvmsg() and reading some specs. Urgent data allows the sender to signal the receiver that “urgent data” of some form has been placed into the packet. The receiver on the other hand must deal with this condition and is forced by himself to handle this condition. If not handled the data is silently ignored. Therefore, it must be negotiated at a higher level that urgent data is transmitted and properly handled at the receiver side.

The urgent bit in the TCP header is set and the Urgent pointer is set to the relevant data. The urgent pointer is a 16 bit value. I wrote “pointer is set to the relevant data”, and here comes the fun into the game! First: the urgent field is an offset that points to the last byte of urgent data. It is up to the receiving application where the urgent data starts. A real world problem is the terminus “points to the last byte of urgent data”. RFC 1011: “The urgent pointer points to the last octet of urgent data (not to the first octet of non-urgent data)”. At data before can be threated as urgent – but this is application level specific.

The problem is the following: the definition of the Urgent Pointer was declared in the RFC 793 – the original TCP standard. Due to some contradictions RFC 1122 updates and correct the behavior (problem was where the urgent pointer points to, the last byte of the urgent data or the byte after the last urgent data). But: all TCP implementations in the wild implement the behavior as specific in RFC 793. The advice by Gont and Yourtchenko is that new applications employing TCP should not use the urgent mechanism at all.