Re: [Lksctp-developers] Re: SCTP config 2.5.70(-bk)

[Jon Grimm <jgrimm2@xxxxxxxxxx>]

Hi Adrian,

	Sorry for a bit of delay... We are away at an SCTP Interoperability 
event.
Adrian Bunk wrote:
> On Mon, Jun 02, 2003 at 09:53:04AM +0200, Margit Schubert-While wrote:
>
>
> > CONFIG_IPV6_SCTP__   is always being set to "y" even though
> > not selected (CONFIG_IPV6 not set)
>
> First, this doesn't do any harm since CONFIG_IPV6_SCTP__ alone doensn't 
> result in anything getting compiled.
> But besides, it seems a bit broken.
>
> From net/sctp/Kconfig:
>
> <--  snip  -->
>
> ...
>
> config IPV6_SCTP__
>         tristate
>         default y if IPV6=n
>         default IPV6 if IPV6
>
> config IP_SCTP
>         tristate "The SCTP Protocol (EXPERIMENTAL)"
>         depends on IPV6_SCTP__
> ...
>
> <--  snip  -->
>
>
> Semantically equivalent is the following for IPV6_SCTP__:
>
> config IPV6_SCTP__
>         tristate
>         default y if IPV6=n || IPV6=y
>         default m if IPV6=m
>
>
> If it was intended to disallow a static IP_SCTP with a modular IPV6 it 
> doesn't work: It's perfectly allowed to set IPV6=n and IP_SCTP=y and 
> later compile and install a modular IPV6 for the same kernel.
Are you sure?  I vaguely remember one of the network structs having 
#ifdef'd fields for v6.   Consequently, if one compiles first without, 
but the tries later compiles/loads ipv6... bad things happen as the 
kernel has a different concept of what the sock is.
> Could someone from the SCTP developers comment on the intentions behind 
> IPV6_SCTP__ ?
Yes.  The intent was to at least discourage a configuration that will 
segfault.
Thanks,
jon

> > Margit
>
> cu
> Adrian