Re: [Lksctp-developers] setting SCTP_EVENTS sockopts fails

[cco <cri...@ip...>]

On Tue, Nov 18, 2008 at 01:15:45PM -0500, Vlad Yasevich wrote:
> cco wrote:
> > On Tue, Nov 18, 2008 at 11:47:42AM -0500, Vlad Yasevich wrote:
> >> cco wrote:
> >>> hi!
> >>>
> >>> any idea why something like:
> >>>
> >>> struct sctp_event_subscribe ev_sub;
> >>> ...
> >>> setsockopt(s, IPPROTO_SCTP, SCTP_EVENTS,
> >>>             &ev_sub, sizeof(ev_sub));
> >>>
> >>> fails with EINVAL?
> >>>
> >>> I am using 2.6.21 / 2.6.22, debian unstable; and I think I saw this
> >>> working before updating debian on those machines...
> >> It might be that the even structures between the kernel and user mismatch.
> >> We used to have a bug that we did very strict size matching that was later
> >> relaxed to allow the the user to have older event structure.
> >>
> >> In any case, you can't have a newer ever structure in userspace with the older
> >> version in the kernel.
> > 
> > cristian: hi again! 
> > 
> > could the mismatch btw. the user space header
> > files/libs and the kernel be the reason why the struct sctp_sndrcvinfo *
> > sinfo from sctp_recvmsg() contains garbagge (at least for ppid) ?
> 
> Never heard of this.  ppid is opaque and the kernel doesn't change the
> byte ordering at all for it.  It's a job of the user application to set
> the byte ordering for it and to do appropriate conversions.

cristian: yes, but if I cannot set the SCTP_EVENTS on the socket (and
thus sctp_data_io_event) I won't get any sctp_sndrcvinfo at
sctp_recvmsg(), right?

thanks again!
bye now!
cristian