[Lksctp-developers] RE: a question about the patch of stale cookie

[<daj...@no...>]

I see.  Thanks  for your clarification.=20

-----Original Message-----
From: ext Jon Grimm [mailto:jg...@us...]
Sent: 25 February 2003 23:38
To: Zhang Dajiang (Nokia-RD/Beijing)
Cc: ard...@in...; sr...@us...; =
Lks...@li...; Huang Hui (Nokia-RD/Beijing)
Subject: Re: a question about the patch of stale cookie




daj...@no... wrote:
>  Hi, ardelle, Jon & sridhar
>=20

Hi back to you!  Good to hear from you again Dajaing.


> I'm porting patch 638235 stale cookie support to our implementation. =
Here are some questions:
> 1. why not combine sctp_sf_cookie_echoed_err with =
sctp_sf_do_5_2_6_stale?


I think they are just split out to allow processing of some other error=20
besides stale cookie.   However, I think that code is a little broke, in =

that its not bumping the INIT error counter for non-stale cookie errors.

> 2. in net/sctp/input.c, our code is as following in case there are =
more than one error cause in ERROR chunk. Do you think so?
> 	if (ch->type =3D=3D SCTP_CID_ERROR) {
> 		/* Need to check the "Stale cookie" ERROR. */
>                 	uint8_t *err_end;
>                =20
>                 	err =3D (sctp_errhdr_t* )((uint8_t *)ch + =
sizeof(sctp_chunkhdr_t));
> 	                do {
> 	                	err_end =3D ((uint8_t *)err)  + =
WORD_ROUND(ntohs(err->length));
>=20
> 	                	if (SCTP_ERROR_STALE_COOKIE =3D=3D err->cause) {
>                                         		goto discard;
> 	                	}
>                        	 } while (err_end < ch_end);
>          		return 0;
> 	}    =20

Yes.  This sounds right.   In a similar vein, this (loop) seems to be=20
needed in sctp_sf_cookie_echo_err & possible other places too?


> 3. in net/sctp/sm_statetable.c, the code in the patch is=20
>=20
>  	/* SCTP_STATE_COOKIE_WAIT */ \
>  	{.fn =3D sctp_sf_not_impl, .name =3D "sctp_sf_not_impl"}, \
>  	/* SCTP_STATE_COOKIE_ECHOED */ \
> -	{.fn =3D sctp_sf_not_impl, .name =3D "sctp_sf_not_impl"}, \
> +	{.fn =3D sctp_sf_cookie_echoed_err, .name =3D =
"sctp_sf_cookie_echoed_err"}, \
>  	/* SCTP_STATE_ESTABLISHED */ \
>  	{.fn =3D sctp_sf_operr_notify, .name =3D "sctp_sf_operr_notify"}, \
>  	/* SCTP_STATE_SHUTDOWN_PENDING */ \=09
>=20
> why not use sctp_sf_operr_notify in case the error cause isn't STALE =
COOKIE but others(e.g.  Out of Resource)?
>=20

How do you know the error cause is not STALE-COOKIE at this point?=20
Regardless, I'm not sure we want to send Op-Errs up the stack as we=20
haven't yet reported as associd to the application.  Not a problem for=20
TCP-style sockets (as they don't use associd.)

However, SCTP_CANT_START_ASSOC notification can report the ERROR cause.=20
  This is something we should be doing, which I don't think we are.

-jon
> BR
> Dajiang
>=20