Herbert Xu píše v Čt 18. 12. 2008 v 08:30 +1100:

On Wed, Dec 17, 2008 at 05:04:48PM +0100, Petr Tesarik wrote:

quoted

This is all nice, but it still does not solve those series of SIGURGs on
the receiving side. My suggestion is to not generate a new SIGURG until
the data for the latest one have arrived. There can be only one byte of
urgent data, so if somebody sends more than one, it cannot be handled by
the receiver, anyway, so that use case is broken and need not be taken
into account.

No we need to send a SIGURG as soon as we enter urgent mode.  The
whole point of urgent mode is to inform the receiver ASAP.  As the
urgent data itself cannot be expedited due to the in-order nature
of TCP, setting the flag and sending the signal is the best we can
do.

That doesn't contradict my suggestion, which is: "Send the SIGURG as
soon as possible, but do not send _more_ SIGURGs until the urgent data
arrives."  I'm going to send a patch to make it clear.

Petr Tesarik