Re: and nicer too - Re: [PATCH] epoll more scalable than poll

[bert hubert]

On Mon, Oct 28, 2002 at 03:45:06PM -0800, John Gardiner Myers wrote:

> As you have amply demonstrated, the current epoll API is error prone. 
> The API should be fixed to test the poll condition and, if necessary, 
> drop an event upon insertion to the set.

That is a semantics change and not an API/ABI change. To reiterate, you
mention the following scenario:

for(;;) {
	nfds = sys_epoll_wait(kdpfd, &pfds, -1);
	for(n = 0; n < nfds; ++n) {
		if((fd = pfds[n].fd) == s) {
                        /* 1: accept client (SYN/SYN|ACK/ACK completed) */
			client = accept(s, (struct sockaddr*)&local, &addrlen);
			if(client < 0){
				perror("accept");
				continue;
			}

			/* 2: packet comes in, client becomes readable  */
			/* 3: registering interest */
			if (sys_epoll_ctl(kdpfd, EP_CTL_ADD, client, POLLIN ) < 0) {
				fprintf(stderr, "sys_epoll set insertion error: fd=%d\n", client);
				return -1;
			}
			
			/* 4: interest only now registered, no edge will be
			   reported, our fd is lost */

			fd = client;
		}
		do_use_fd(fd);
	}
}                                                                                                                     

There are lots of ways to solve this, I bet Davide knows best. Perhaps it is
solved already, you can't tell from only studying the API, the problem isn't
intrinsic to it.  

An easy solution is to have sys_epoll_ctl check if there is there is data
ready and make sure there is an edge to report in that case to the next call
of sys_epoll_ctl().

Regards,

bert

-- 
http://www.PowerDNS.com          Versatile DNS Software & Services
http://lartc.org           Linux Advanced Routing & Traffic Control HOWTO