Re: Regarding to your linux kernel CL

[Herbert Xu <herbert@gondor.apana.org.au>]

On Wed, Oct 06, 2010 at 10:02:56AM +0300, Timo Teräs wrote:
>
> What's the problem here? sk_dst_check not honoring if dst->obsolete>0 ?
> Sounds like the sk_dst_check was buggy in the first place.

Well the problem is that before we changed ip4_dst_check, everything
worked properly.  With IPv6, whenever a route is released, the serial
number is always updated accordingly.  This means that ip6_dst_check
will always return NULL when obsolete > 1.

The old ip4_dst_check also satisfied this requirement since it always
returns NULL.

> Looks like there's still some code around that does not do what the
> obsolete field has been used for a long time.
>   obsolete =  0, dst entry is ok
>   obsolete = -1, you need to call ops->check for this entry
>   obsolete >  0, this entry is invalid
> 
> So net/core/sock.c needs fixing. Just if we should change dst_check()
> too, I'm not sure.
> 
> Should we fix sk_dst_check to use dst_check(), and dst_check() to check
> for dst->obsolete>0 ?

Yes this should work too.  However, I was never totally happy with
this new dst->obsolete logic which means that we're doing an
indirect call for every single packet which almost always does
nothing.

Perhaps we should move the genid/cookie logic into the dst so that
we can eliminate the dst->check call or at least make it a lot less
frequent.

Cheers,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt