On Tue, 18 Dec 2001, Alfred Perlstein wrote:

> > * Bruce Evans <[EMAIL PROTECTED]> [011217 07:10] wrote:
> > > I'm not happy with frobbing the socket state.  I suggest frobbing the
> > > events mask instead.  Either use a flag to tell sopoll() to ignore
> > > SS_CANTRCVMORE, or use new events POLLIN_IGNORE_EOF and
> > > POLLRDNORM_IGNORE_EOF (convert the userland POLLIN/POLLRDNORM to these
> > > and change sopoll() to support them).
>
> Wait, please clarify, are you suggesting that POLLIN_IGNORE_EOF
> and POLLRDNORM_IGNORE_EOF must be used by userland applications?
> Or would it be an internal flag for the kernel so that here you'd
> see something like:
>
> filetmp.f_data = (caddr_t)ap->a_vp->v_fifoinfo->fi_readsock;
> if (filetmp.f_data) {
>       if (ap->a_events & ~POLLIN)
>               ap->a_events |= POLLIN_IGNORE_EOF;

I think it should be more like:

        if (ap->a_events & POLLIN)
                ap->a_events = (ap->a_events & ~POLLIN) | POLLIN_IGNORE_EOF;

We can't pass in flags that we don't really care about (POLLIN in
this case) since they will cause side effects (selrecord()...).

I might actually use a new variable instead of changing `ap'.

>       /* same for POLLRDNORM */
>       revents |= soo_poll(&filetmp, ap->a_events, ap->a_cred,
>                       ap->a_td);

Add something like:

        if (revents & POLLIN_IGNORE_EOF)
                revents = (revents & ~POLLIN_IGNORE_EOF) | POLLIN;

> }
>
> Or that the userland will just OR in POLLIN_IGNORE_EOF before calling
> poll(2).
>
> I'm not sure if I like the latter way of fixing things.

Userland might want to do this someday.  I would just make it easy for it
to do so.

Bruce


To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-net" in the body of the message

Reply via email to