ipv6: Create percpu rt6_info

After the patch
'ipv6: Only create RTF_CACHE routes after encountering pmtu exception',
we need to compensate the performance hit (bouncing dst->__refcnt).

Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Cc: Hannes Frederic Sowa <hannes@stressinduktion.org>
Cc: Steffen Klassert <steffen.klassert@secunet.com>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/net/ip6_fib.h

@@ -124,6 +124,7 @@ struct rt6_info {
 	struct uncached_list		*rt6i_uncached_list;
 
 	struct inet6_dev		*rt6i_idev;
+	struct rt6_info * __percpu	*rt6i_pcpu;
 
 	u32				rt6i_metric;
 	u32				rt6i_pmtu;
@@ -164,7 +165,7 @@ static inline void rt6_update_expires(struct rt6_info *rt0, int timeout)
 
 static inline u32 rt6_get_cookie(const struct rt6_info *rt)
 {
-	if (unlikely(rt->dst.flags & DST_NOCACHE))
+	if (rt->rt6i_flags & RTF_PCPU || unlikely(rt->dst.flags & DST_NOCACHE))
 		rt = (struct rt6_info *)(rt->dst.from);
 
 	return rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;

include/uapi/linux/ipv6_route.h

@@ -34,6 +34,7 @@
 #define RTF_PREF(pref)	((pref) << 27)
 #define RTF_PREF_MASK	0x18000000
 
+#define RTF_PCPU	0x40000000
 #define RTF_LOCAL	0x80000000
 
 

net/ipv6/ip6_fib.c

@@ -154,10 +154,32 @@ static void node_free(struct fib6_node *fn)
 	kmem_cache_free(fib6_node_kmem, fn);
 }
 
+static void rt6_free_pcpu(struct rt6_info *non_pcpu_rt)
+{
+	int cpu;
+
+	if (!non_pcpu_rt->rt6i_pcpu)
+		return;
+
+	for_each_possible_cpu(cpu) {
+		struct rt6_info **ppcpu_rt;
+		struct rt6_info *pcpu_rt;
+
+		ppcpu_rt = per_cpu_ptr(non_pcpu_rt->rt6i_pcpu, cpu);
+		pcpu_rt = *ppcpu_rt;
+		if (pcpu_rt) {
+			dst_free(&pcpu_rt->dst);
+			*ppcpu_rt = NULL;
+		}
+	}
+}
+
 static void rt6_release(struct rt6_info *rt)
 {
-	if (atomic_dec_and_test(&rt->rt6i_ref))
+	if (atomic_dec_and_test(&rt->rt6i_ref)) {
+		rt6_free_pcpu(rt);
 		dst_free(&rt->dst);
+	}
 }
 
 static void fib6_link_table(struct net *net, struct fib6_table *tb)

net/ipv6/route.c

@@ -165,11 +165,18 @@ static void rt6_uncached_list_flush_dev(struct net *net, struct net_device *dev)
 	}
 }
 
+static u32 *rt6_pcpu_cow_metrics(struct rt6_info *rt)
+{
+	return dst_metrics_write_ptr(rt->dst.from);
+}
+
 static u32 *ipv6_cow_metrics(struct dst_entry *dst, unsigned long old)
 {
 	struct rt6_info *rt = (struct rt6_info *)dst;
 
-	if (rt->rt6i_flags & RTF_CACHE)
+	if (rt->rt6i_flags & RTF_PCPU)
+		return rt6_pcpu_cow_metrics(rt);
+	else if (rt->rt6i_flags & RTF_CACHE)
 		return NULL;
 	else
 		return dst_cow_metrics_generic(dst, old);
@@ -309,10 +316,10 @@ static const struct rt6_info ip6_blk_hole_entry_template = {
 #endif
 
 /* allocate dst with ip6_dst_ops */
-static inline struct rt6_info *ip6_dst_alloc(struct net *net,
-					     struct net_device *dev,
-					     int flags,
-					     struct fib6_table *table)
+static struct rt6_info *__ip6_dst_alloc(struct net *net,
+					struct net_device *dev,
+					int flags,
+					struct fib6_table *table)
 {
 	struct rt6_info *rt = dst_alloc(&net->ipv6.ip6_dst_ops, dev,
 					0, DST_OBSOLETE_FORCE_CHK, flags);
@@ -327,6 +334,34 @@ static inline struct rt6_info *ip6_dst_alloc(struct net *net,
 	return rt;
 }
 
+static struct rt6_info *ip6_dst_alloc(struct net *net,
+				      struct net_device *dev,
+				      int flags,
+				      struct fib6_table *table)
+{
+	struct rt6_info *rt = __ip6_dst_alloc(net, dev, flags, table);
+
+	if (rt) {
+		rt->rt6i_pcpu = alloc_percpu_gfp(struct rt6_info *, GFP_ATOMIC);
+		if (rt->rt6i_pcpu) {
+			int cpu;
+
+			for_each_possible_cpu(cpu) {
+				struct rt6_info **p;
+
+				p = per_cpu_ptr(rt->rt6i_pcpu, cpu);
+				/* no one shares rt */
+				*p =  NULL;
+			}
+		} else {
+			dst_destroy((struct dst_entry *)rt);
+			return NULL;
+		}
+	}
+
+	return rt;
+}
+
 static void ip6_dst_destroy(struct dst_entry *dst)
 {
 	struct rt6_info *rt = (struct rt6_info *)dst;
@@ -335,6 +370,9 @@ static void ip6_dst_destroy(struct dst_entry *dst)
 
 	dst_destroy_metrics_generic(dst);
 
+	if (rt->rt6i_pcpu)
+		free_percpu(rt->rt6i_pcpu);
+
 	rt6_uncached_list_del(rt);
 
 	idev = rt->rt6i_idev;
@@ -912,11 +950,11 @@ static struct rt6_info *ip6_rt_cache_alloc(struct rt6_info *ort,
 	 *	Clone the route.
 	 */
 
-	if (ort->rt6i_flags & RTF_CACHE)
+	if (ort->rt6i_flags & (RTF_CACHE | RTF_PCPU))
 		ort = (struct rt6_info *)ort->dst.from;
 
-	rt = ip6_dst_alloc(dev_net(ort->dst.dev), ort->dst.dev,
-			   0, ort->rt6i_table);
+	rt = __ip6_dst_alloc(dev_net(ort->dst.dev), ort->dst.dev,
+			     0, ort->rt6i_table);
 
 	if (!rt)
 		return NULL;
@@ -943,6 +981,54 @@ static struct rt6_info *ip6_rt_cache_alloc(struct rt6_info *ort,
 	return rt;
 }
 
+static struct rt6_info *ip6_rt_pcpu_alloc(struct rt6_info *rt)
+{
+	struct rt6_info *pcpu_rt;
+
+	pcpu_rt = __ip6_dst_alloc(dev_net(rt->dst.dev),
+				  rt->dst.dev, rt->dst.flags,
+				  rt->rt6i_table);
+
+	if (!pcpu_rt)
+		return NULL;
+	ip6_rt_copy_init(pcpu_rt, rt);
+	pcpu_rt->rt6i_protocol = rt->rt6i_protocol;
+	pcpu_rt->rt6i_flags |= RTF_PCPU;
+	return pcpu_rt;
+}
+
+/* It should be called with read_lock_bh(&tb6_lock) acquired */
+static struct rt6_info *rt6_get_pcpu_route(struct rt6_info *rt)
+{
+	struct rt6_info *pcpu_rt, *prev, **p;
+
+	p = this_cpu_ptr(rt->rt6i_pcpu);
+	pcpu_rt = *p;
+
+	if (pcpu_rt)
+		goto done;
+
+	pcpu_rt = ip6_rt_pcpu_alloc(rt);
+	if (!pcpu_rt) {
+		struct net *net = dev_net(rt->dst.dev);
+
+		pcpu_rt = net->ipv6.ip6_null_entry;
+		goto done;
+	}
+
+	prev = cmpxchg(p, NULL, pcpu_rt);
+	if (prev) {
+		/* If someone did it before us, return prev instead */
+		dst_destroy(&pcpu_rt->dst);
+		pcpu_rt = prev;
+	}
+
+done:
+	dst_hold(&pcpu_rt->dst);
+	rt6_dst_from_metrics_check(pcpu_rt);
+	return pcpu_rt;
+}
+
 static struct rt6_info *ip6_pol_route(struct net *net, struct fib6_table *table, int oif,
 				      struct flowi6 *fl6, int flags)
 {
@@ -975,11 +1061,13 @@ redo_rt6_select:
 		}
 	}
 
-	dst_use(&rt->dst, jiffies);
-	read_unlock_bh(&table->tb6_lock);
 
 	if (rt == net->ipv6.ip6_null_entry || (rt->rt6i_flags & RTF_CACHE)) {
-		goto done;
+		dst_use(&rt->dst, jiffies);
+		read_unlock_bh(&table->tb6_lock);
+
+		rt6_dst_from_metrics_check(rt);
+		return rt;
 	} else if (unlikely((fl6->flowi6_flags & FLOWI_FLAG_KNOWN_NH) &&
 			    !(rt->rt6i_flags & RTF_GATEWAY))) {
 		/* Create a RTF_CACHE clone which will not be
@@ -990,6 +1078,9 @@ redo_rt6_select:
 
 		struct rt6_info *uncached_rt;
 
+		dst_use(&rt->dst, jiffies);
+		read_unlock_bh(&table->tb6_lock);
+
 		uncached_rt = ip6_rt_cache_alloc(rt, &fl6->daddr, NULL);
 		dst_release(&rt->dst);
 
@@ -997,13 +1088,22 @@ redo_rt6_select:
 			rt6_uncached_list_add(uncached_rt);
 		else
 			uncached_rt = net->ipv6.ip6_null_entry;
+
 		dst_hold(&uncached_rt->dst);
 		return uncached_rt;
-	}
 
-done:
-	rt6_dst_from_metrics_check(rt);
-	return rt;
+	} else {
+		/* Get a percpu copy */
+
+		struct rt6_info *pcpu_rt;
+
+		rt->dst.lastuse = jiffies;
+		rt->dst.__use++;
+		pcpu_rt = rt6_get_pcpu_route(rt);
+		read_unlock_bh(&table->tb6_lock);
+
+		return pcpu_rt;
+	}
 }
 
 static struct rt6_info *ip6_pol_route_input(struct net *net, struct fib6_table *table,
@@ -1147,7 +1247,7 @@ static struct dst_entry *ip6_dst_check(struct dst_entry *dst, u32 cookie)
 
 	rt6_dst_from_metrics_check(rt);
 
-	if (unlikely(dst->flags & DST_NOCACHE))
+	if ((rt->rt6i_flags & RTF_PCPU) || unlikely(dst->flags & DST_NOCACHE))
 		return rt6_dst_from_check(rt, cookie);
 	else
 		return rt6_check(rt, cookie);