On Tue, 16 Aug 2005, Herbert Xu wrote: > On Thu, Aug 04, 2005 at 05:48:17PM -0700, Arthur Kepner wrote: > > > > @@ -366,6 +376,9 @@ static struct ipq *ip_frag_create(unsign > > ..... > > + qp->peer = sysctl_ipfrag_max_dist ? inet_getpeer(iph->saddr, 1) : NULL; > > + if (qp->peer) > > + qp->rid = atomic_read(&qp->peer->rid); > > This fix is not needed. The rid was set below in ip_frag_too_far which > is called from ip_frag_queue.
I think this and your next comment are related, so please see my response below. > > > +static inline int ip_frag_too_far(struct ipq *qp) > > ..... > > + start = ++qp->rid; > > + end = atomic_inc_return(&peer->rid); > > I'm not sure I understand the benefit of this change either. Your patch http://marc.theaimsgroup.com/?l=linux-netdev&m=112281936522415&w=2 had: +static inline int ip_frag_too_far(struct ipq *qp) + .... + unsigned int max = sysctl_ipfrag_max_dist; + .... + start = qp->rid; + end = atomic_inc_return(&peer->rid); + qp->rid = end; + return qp->fragments && (unsigned int)(end - start) >= max; So, your patch would accept up to sysctl_ipfrag_max_dist "disorder" between calls to ip_frag_too_far() (for a particluar fragment queue). (I'm using "disorder" loosely, but I think it's clear enough what is meant?) With my change it would accept up to sysctl_ipfrag_max_dist "disorder" over the lifetime of a fragment queue. The latter is the behavior that I intended. But your original may be better, especially if there are many fragment queues with a common source. > ...... > > +static int ip_frag_reinit(struct ipq *qp) > > > + if (sysctl_ipfrag_max_dist) { > > + if (qp->peer == NULL) { > > + qp->peer = inet_getpeer(qp->saddr, 1); > > + } > > + if (qp->peer) { > > + qp->rid = atomic_read(&qp->peer->rid); > > + } > > + } > > This is unnecessary since we're taking over an existing qp > whose parameters are exactly the same as the new fragment. > This is similar to the first comment (about ip_frag_create()). And the difference is to do with how I changed the way qp->rid is updated in ip_frag_too_far(). -- Arthur - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
