[PATCH v3] ipv6: Not to probe neighbourless routes

[PATCH v3] ipv6: Not to probe neighbourless routes

[Cheng Lin]

From: Cheng Lin <cheng.lin130@zte.com.cn>

Originally, Router Reachability Probing require a neighbour entry
existed. Commit 2152caea7196 ("ipv6: Do not depend on rt->n in
rt6_probe().") removed the requirement for a neighbour entry. And
commit f547fac624be ("ipv6: rate-limit probes for neighbourless
routes") adds rate-limiting for neighbourless routes.

And, the Neighbor Discovery for IP version 6 (IPv6)(rfc4861) says,
"
7.2.5.  Receipt of Neighbor Advertisements

When a valid Neighbor Advertisement is received (either solicited or
unsolicited), the Neighbor Cache is searched for the target's entry.
If no entry exists, the advertisement SHOULD be silently discarded.
There is no need to create an entry if none exists, since the
recipient has apparently not initiated any communication with the
target.
".

In rt6_probe(), just a Neighbor Solicitation message are transmited.
When receiving a Neighbor Advertisement, the node does nothing in a
Neighborless condition.

Not sure it's needed to create a neighbor entry in Router
Reachability Probing. And the Original way may be the right way.

This patch recover the requirement for a neighbour entry.

Signed-off-by: Cheng Lin <cheng.lin130@zte.com.cn>
---
 include/net/ip6_fib.h | 5 -----
 net/ipv6/route.c      | 6 +-----
 2 files changed, 1 insertion(+), 10 deletions(-)

diff --git a/include/net/ip6_fib.h b/include/net/ip6_fib.h
index 4b5656c..8c2e022 100644
--- a/include/net/ip6_fib.h
+++ b/include/net/ip6_fib.h
@@ -124,11 +124,6 @@ struct rt6_exception {
 
 struct fib6_nh {
 	struct fib_nh_common	nh_common;
-
-#ifdef CONFIG_IPV6_ROUTER_PREF
-	unsigned long		last_probe;
-#endif
-
 	struct rt6_info * __percpu *rt6i_pcpu;
 	struct rt6_exception_bucket __rcu *rt6i_exception_bucket;
 };
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index fd059e0..1839dd7 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -639,12 +639,12 @@ static void rt6_probe(struct fib6_nh *fib6_nh)
 	nh_gw = &fib6_nh->fib_nh_gw6;
 	dev = fib6_nh->fib_nh_dev;
 	rcu_read_lock_bh();
-	idev = __in6_dev_get(dev);
 	neigh = __ipv6_neigh_lookup_noref(dev, nh_gw);
 	if (neigh) {
 		if (neigh->nud_state & NUD_VALID)
 			goto out;
 
+		idev = __in6_dev_get(dev);
 		write_lock(&neigh->lock);
 		if (!(neigh->nud_state & NUD_VALID) &&
 		    time_after(jiffies,
@@ -654,13 +654,9 @@ static void rt6_probe(struct fib6_nh *fib6_nh)
 				__neigh_set_probe_once(neigh);
 		}
 		write_unlock(&neigh->lock);
-	} else if (time_after(jiffies, fib6_nh->last_probe +
-				       idev->cnf.rtr_probe_interval)) {
-		work = kmalloc(sizeof(*work), GFP_ATOMIC);
 	}
 
 	if (work) {
-		fib6_nh->last_probe = jiffies;
 		INIT_WORK(&work->work, rt6_probe_deferred);
 		work->target = *nh_gw;
 		dev_hold(dev);
-- 
1.8.3.1

Re: [PATCH v3] ipv6: Not to probe neighbourless routes

[David Miller]

From: Cheng Lin <wang.yi59@zte.com.cn>
Date: Fri, 30 Aug 2019 14:11:16 +0800

> Originally, Router Reachability Probing require a neighbour entry
> existed. Commit 2152caea7196 ("ipv6: Do not depend on rt->n in
> rt6_probe().") removed the requirement for a neighbour entry. And
> commit f547fac624be ("ipv6: rate-limit probes for neighbourless
> routes") adds rate-limiting for neighbourless routes.

I am not going to apply this patch.

The reason we handle neighbourless routes is because due to the
disconnect between routes and neighbour entries, we would lose
information with your suggested change.

Originally, all routes held a reference to a neighbour entry.
Therefore we'd always have a neigh entry for any neigh message
matching a route.

But these two object pools (routes and neigh entries) are completely
disconnected.  We only look up a neigh entry when sending a packet
on behalf of a route.

Therfore, neigh entries can be purged arbitrarily even if hundreds of
routes refer to them.  And this means it is very important to accept
and process probes even for neighbourless routes.

I would also not recommend, in the future, reading RFC requirements
literally without taking into consideration the details of Linux's
specific implementation of ipv6 routing and neighbours.

Thank you.