On Fri, Mar 21, 2008 at 8:57 AM, Kurt Miller <[EMAIL PROTECTED]> wrote: > On Friday 21 March 2008 10:47:27 am Kurt Miller wrote: > > On Friday 21 March 2008 6:25:59 am Philip Guenther wrote: ... > > > As a gross kludge, I think things would "work" if you added the > > > following to the code called by the child after the fork. > > > sleep(1); /* make sure the parent has a chance to exit */ > > > fcntl(sockfd, F_SETFL, fcntl(sockfd, F_GETFL) & ~O_NONBLOCK); > > > > > > (With the correct 'sockfd' variable, of course). That'll bring the > > > kernel's O_NONBLOCK flag on the socket in sync with what libpthread > > > thinks it should be. > > > > To avoid the race in the child (and the sleep() call), the parrent can set > the > > fd's to non-blocking *before* the fork() and the child can set them back to > > blocking directly after the fork. Obviously not ideal but it will > work-around > > the problem effectively.
Yeah, that's better than my suggestion. > After some more thought, a simpler solution would be to just have the parrent > set the fd's to non-blocking directly before the exit() call. Unfortunately, that requires having the sockfd on hand in the function that calls fork()+exit(), so I suspect it'll be more intrusive than flipping the fd before and after. Philip Guenther
