[net-next v1] net: ipv6: avoid overhead when no custom FIB rules are installed

Subsystems: networking [general], networking [ipv4/ipv6], the rest

7 messages, 4 authors, 2017-08-09 · open the first message on its own page

[net-next v1] net: ipv6: avoid overhead when no custom FIB rules are installed

From: Vincent Bernat <hidden>
Date: 2017-08-08 11:51:39

If the user hasn't installed any custom rules, don't go through the
whole FIB rules layer. This is pretty similar to f4530fa574df (ipv4:
Avoid overhead when no custom FIB rules are installed).

Using a micro-benchmark module [1], timing ip6_route_output() using
get_cycles(), on my laptop with performance governor (so, subject to
some caution due to thermal throttling), with 40,000 routes in the main
routing table, before this patch:

    min=603 max=12477 count=627 average=1926 95th=4518 90th=3661 50th=1600 mad=724
    table=254 avgdepth=21.8 maxdepth=39
        value │                         ┊                            count
          600 │▒▒▒▒▒▒▒▒▒▒                                              134
          860 │▒▒▒▒▒▒▒░░░░░░░░░░                                        91
         1120 │▒▒▒░░░░░░░░░░░░░░░░░░                                    39
         1380 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░                                58
         1640 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░                           57
         1900 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                       58
         2160 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                   46
         2420 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                 26
         2680 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░               27
         2940 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              15
         3200 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              8
         3460 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             8
         3720 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           12
         3980 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            9
         4240 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           7
         4500 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           6

After:

    min=486 max=12390 count=627 average=1715 95th=4365 90th=3352 50th=1324 mad=631
    table=254 avgdepth=21.8 maxdepth=39
        value │                         ┊                            count
          480 │▒▒▒▒▒▒▒▒▒▒▒▒                                            153
          730 │▒▒▒▒▒▒▒░░░░░░░░░░░░                                      88
          980 │▒▒▒▒░░░░░░░░░░░░░░░░░░░                                  57
         1230 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░                             56
         1480 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░                          43
         1730 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                      50
         1980 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                   31
         2230 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                  23
         2480 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                15
         2730 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░               23
         2980 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             20
         3230 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            14
         3480 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            4
         3730 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            8
         3980 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           9
         4230 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           4

At the frequency of the laptop during the bench (~ 2.3 GHz), this is
about a 130 ns difference on the median value and 50 ns on the minimal
value.

A next step would be to collapse local and main tables, as in
0ddcf43d5d4a (ipv4: FIB Local/MAIN table collapse).

[1]: https://github.com/vincentbernat/network-lab/blob/master/lab-routes-ipv6/kbench_mod.c

Signed-off-by: Vincent Bernat <redacted>
---
 include/net/netns/ipv6.h |  1 +
 net/ipv6/fib6_rules.c    | 34 ++++++++++++++++++++++++----------
 net/ipv6/route.c         |  1 +
 3 files changed, 26 insertions(+), 10 deletions(-)
diff --git a/include/net/netns/ipv6.h b/include/net/netns/ipv6.h
index abdf3b40303b..0e50bf3ed097 100644
--- a/include/net/netns/ipv6.h
+++ b/include/net/netns/ipv6.h
@@ -65,6 +65,7 @@ struct netns_ipv6 {
 	unsigned int		 ip6_rt_gc_expire;
 	unsigned long		 ip6_rt_last_gc;
 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
+	bool			 fib6_has_custom_rules;
 	struct rt6_info         *ip6_prohibit_entry;
 	struct rt6_info         *ip6_blk_hole_entry;
 	struct fib6_table       *fib6_local_tbl;
diff --git a/net/ipv6/fib6_rules.c b/net/ipv6/fib6_rules.c
index 2f29e4e33bd3..693c27ede40e 100644
--- a/net/ipv6/fib6_rules.c
+++ b/net/ipv6/fib6_rules.c
@@ -63,19 +63,32 @@ unsigned int fib6_rules_seq_read(struct net *net)
 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 				   int flags, pol_lookup_t lookup)
 {
-	struct fib_lookup_arg arg = {
-		.lookup_ptr = lookup,
-		.flags = FIB_LOOKUP_NOREF,
-	};
-
 	/* update flow if oif or iif point to device enslaved to l3mdev */
 	l3mdev_update_flow(net, flowi6_to_flowi(fl6));
 
-	fib_rules_lookup(net->ipv6.fib6_rules_ops,
-			 flowi6_to_flowi(fl6), flags, &arg);
-
-	if (arg.result)
-		return arg.result;
+	if (net->ipv6.fib6_has_custom_rules) {
+		struct fib_lookup_arg arg = {
+			.lookup_ptr = lookup,
+			.flags = FIB_LOOKUP_NOREF,
+		};
+
+		fib_rules_lookup(net->ipv6.fib6_rules_ops,
+				 flowi6_to_flowi(fl6), flags, &arg);
+
+		if (arg.result)
+			return arg.result;
+	} else {
+		struct rt6_info *rt;
+
+		rt = lookup(net, net->ipv6.fib6_local_tbl, fl6, flags);
+		if (rt != net->ipv6.ip6_null_entry && rt->dst.error != -EAGAIN)
+			return &rt->dst;
+		ip6_rt_put(rt);
+		rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
+		if (rt->dst.error != -EAGAIN)
+			return &rt->dst;
+		ip6_rt_put(rt);
+	}
 
 	dst_hold(&net->ipv6.ip6_null_entry->dst);
 	return &net->ipv6.ip6_null_entry->dst;
@@ -245,6 +258,7 @@ static int fib6_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 	rule6->dst.plen = frh->dst_len;
 	rule6->tclass = frh->tos;
 
+	net->ipv6.fib6_has_custom_rules = true;
 	err = 0;
 errout:
 	return err;
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index aba07fce67fb..7ecbe5eb19f8 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -3934,6 +3934,7 @@ static int __net_init ip6_route_net_init(struct net *net)
 			 ip6_template_metrics, true);
 
 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
+	net->ipv6.fib6_has_custom_rules = false;
 	net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
 					       sizeof(*net->ipv6.ip6_prohibit_entry),
 					       GFP_KERNEL);
-- 
2.13.3

Re: [net-next v1] net: ipv6: avoid overhead when no custom FIB rules are installed

From: Jiri Pirko <jiri@resnulli.us>
Date: 2017-08-08 12:51:35

Tue, Aug 08, 2017 at 01:51:21PM CEST, vincent@bernat.im wrote:
If the user hasn't installed any custom rules, don't go through the
whole FIB rules layer. This is pretty similar to f4530fa574df (ipv4:
Avoid overhead when no custom FIB rules are installed).

Using a micro-benchmark module [1], timing ip6_route_output() using
get_cycles(), on my laptop with performance governor (so, subject to
some caution due to thermal throttling), with 40,000 routes in the main
routing table, before this patch:

   min=603 max=12477 count=627 average=1926 95th=4518 90th=3661 50th=1600 mad=724
   table=254 avgdepth=21.8 maxdepth=39
       value │                         ┊                            count
         600 │▒▒▒▒▒▒▒▒▒▒                                              134
         860 │▒▒▒▒▒▒▒░░░░░░░░░░                                        91
        1120 │▒▒▒░░░░░░░░░░░░░░░░░░                                    39
        1380 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░                                58
        1640 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░                           57
        1900 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                       58
        2160 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                   46
        2420 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                 26
        2680 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░               27
        2940 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              15
        3200 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              8
        3460 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             8
        3720 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           12
        3980 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            9
        4240 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           7
        4500 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           6

After:

   min=486 max=12390 count=627 average=1715 95th=4365 90th=3352 50th=1324 mad=631
   table=254 avgdepth=21.8 maxdepth=39
       value │                         ┊                            count
         480 │▒▒▒▒▒▒▒▒▒▒▒▒                                            153
         730 │▒▒▒▒▒▒▒░░░░░░░░░░░░                                      88
         980 │▒▒▒▒░░░░░░░░░░░░░░░░░░░                                  57
        1230 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░                             56
        1480 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░                          43
        1730 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                      50
        1980 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                   31
        2230 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                  23
        2480 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                15
        2730 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░               23
        2980 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             20
        3230 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            14
        3480 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            4
        3730 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            8
        3980 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           9
        4230 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           4
Cool graphs!

At the frequency of the laptop during the bench (~ 2.3 GHz), this is
about a 130 ns difference on the median value and 50 ns on the minimal
value.

A next step would be to collapse local and main tables, as in
0ddcf43d5d4a (ipv4: FIB Local/MAIN table collapse).

[1]: https://github.com/vincentbernat/network-lab/blob/master/lab-routes-ipv6/kbench_mod.c

Signed-off-by: Vincent Bernat <redacted>
Reviewed-by: Jiri Pirko <redacted>

Re: [net-next v1] net: ipv6: avoid overhead when no custom FIB rules are installed

From: David Ahern <hidden>
Date: 2017-08-08 14:46:27

On 8/8/17 5:51 AM, Vincent Bernat wrote:
quoted hunk
diff --git a/net/ipv6/fib6_rules.c b/net/ipv6/fib6_rules.c
index 2f29e4e33bd3..693c27ede40e 100644
--- a/net/ipv6/fib6_rules.c
+++ b/net/ipv6/fib6_rules.c
@@ -63,19 +63,32 @@ unsigned int fib6_rules_seq_read(struct net *net)
 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 				   int flags, pol_lookup_t lookup)
 {
-	struct fib_lookup_arg arg = {
-		.lookup_ptr = lookup,
-		.flags = FIB_LOOKUP_NOREF,
-	};
-
 	/* update flow if oif or iif point to device enslaved to l3mdev */
 	l3mdev_update_flow(net, flowi6_to_flowi(fl6));
The l3mdev_update_flow can be moved to the has_custom_rules block.
l3mdev requires FIB rules for the lookups to work, so no rules means no
l3mdev configured.

Rest looks good to me.

quoted hunk
 
-	fib_rules_lookup(net->ipv6.fib6_rules_ops,
-			 flowi6_to_flowi(fl6), flags, &arg);
-
-	if (arg.result)
-		return arg.result;
+	if (net->ipv6.fib6_has_custom_rules) {
+		struct fib_lookup_arg arg = {
+			.lookup_ptr = lookup,
+			.flags = FIB_LOOKUP_NOREF,
+		};
+
+		fib_rules_lookup(net->ipv6.fib6_rules_ops,
+				 flowi6_to_flowi(fl6), flags, &arg);
+
+		if (arg.result)
+			return arg.result;
+	} else {
+		struct rt6_info *rt;
+
+		rt = lookup(net, net->ipv6.fib6_local_tbl, fl6, flags);
+		if (rt != net->ipv6.ip6_null_entry && rt->dst.error != -EAGAIN)
+			return &rt->dst;
+		ip6_rt_put(rt);
+		rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
+		if (rt->dst.error != -EAGAIN)
+			return &rt->dst;
+		ip6_rt_put(rt);
+	}
 
 	dst_hold(&net->ipv6.ip6_null_entry->dst);
 	return &net->ipv6.ip6_null_entry->dst;

Re: [net-next v1] net: ipv6: avoid overhead when no custom FIB rules are installed

From: Vincent Bernat <hidden>
Date: 2017-08-08 14:57:41

 ❦  8 août 2017 08:46 -0600, David Ahern [off-list ref] :
quoted
diff --git a/net/ipv6/fib6_rules.c b/net/ipv6/fib6_rules.c
index 2f29e4e33bd3..693c27ede40e 100644
--- a/net/ipv6/fib6_rules.c
+++ b/net/ipv6/fib6_rules.c
@@ -63,19 +63,32 @@ unsigned int fib6_rules_seq_read(struct net *net)
 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 				   int flags, pol_lookup_t lookup)
 {
-	struct fib_lookup_arg arg = {
-		.lookup_ptr = lookup,
-		.flags = FIB_LOOKUP_NOREF,
-	};
-
 	/* update flow if oif or iif point to device enslaved to l3mdev */
 	l3mdev_update_flow(net, flowi6_to_flowi(fl6));
The l3mdev_update_flow can be moved to the has_custom_rules block.
l3mdev requires FIB rules for the lookups to work, so no rules means no
l3mdev configured.

Rest looks good to me.
I suspected that it could be moved. I'll update the patch tomorrow.
-- 
Choose variable names that won't be confused.
            - The Elements of Programming Style (Kernighan & Plauger)

[PATCH net-next v2] net: ipv6: avoid overhead when no custom FIB rules are installed

From: Vincent Bernat <hidden>
Date: 2017-08-08 18:24:03

If the user hasn't installed any custom rules, don't go through the
whole FIB rules layer. This is pretty similar to f4530fa574df (ipv4:
Avoid overhead when no custom FIB rules are installed).

Using a micro-benchmark module [1], timing ip6_route_output() with
get_cycles(), with 40,000 routes in the main routing table, before this
patch:

    min=606 max=12911 count=627 average=1959 95th=4903 90th=3747 50th=1602 mad=821
    table=254 avgdepth=21.8 maxdepth=39
    value │                         ┊                            count
      600 │▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒                                         199
      880 │▒▒▒░░░░░░░░░░░░░░░░                                      43
     1160 │▒▒▒░░░░░░░░░░░░░░░░░░░░                                  48
     1440 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░                               43
     1720 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░                          59
     2000 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                      50
     2280 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                    26
     2560 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                  31
     2840 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░               28
     3120 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              17
     3400 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             17
     3680 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             8
     3960 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           11
     4240 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            6
     4520 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           6
     4800 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           9

After:

    min=544 max=11687 count=627 average=1776 95th=4546 90th=3585 50th=1227 mad=565
    table=254 avgdepth=21.8 maxdepth=39
    value │                         ┊                            count
      540 │▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒                                        201
      800 │▒▒▒▒▒░░░░░░░░░░░░░░░░                                    63
     1060 │▒▒▒▒▒░░░░░░░░░░░░░░░░░░░░░                               68
     1320 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░                            39
     1580 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                         32
     1840 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                       32
     2100 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                    34
     2360 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                 33
     2620 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░               26
     2880 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              22
     3140 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              9
     3400 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             8
     3660 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             9
     3920 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            8
     4180 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           8
     4440 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           8

At the frequency of the host during the bench (~ 3.7 GHz), this is
about a 100 ns difference on the median value.

A next step would be to collapse local and main tables, as in
0ddcf43d5d4a (ipv4: FIB Local/MAIN table collapse).

[1]: https://github.com/vincentbernat/network-lab/blob/master/lab-routes-ipv6/kbench_mod.c

Signed-off-by: Vincent Bernat <redacted>
Reviewed-by: Jiri Pirko <redacted>
---
 include/net/netns/ipv6.h |  1 +
 net/ipv6/fib6_rules.c    | 40 +++++++++++++++++++++++++++-------------
 net/ipv6/route.c         |  1 +
 3 files changed, 29 insertions(+), 13 deletions(-)
diff --git a/include/net/netns/ipv6.h b/include/net/netns/ipv6.h
index abdf3b40303b..0e50bf3ed097 100644
--- a/include/net/netns/ipv6.h
+++ b/include/net/netns/ipv6.h
@@ -65,6 +65,7 @@ struct netns_ipv6 {
 	unsigned int		 ip6_rt_gc_expire;
 	unsigned long		 ip6_rt_last_gc;
 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
+	bool			 fib6_has_custom_rules;
 	struct rt6_info         *ip6_prohibit_entry;
 	struct rt6_info         *ip6_blk_hole_entry;
 	struct fib6_table       *fib6_local_tbl;
diff --git a/net/ipv6/fib6_rules.c b/net/ipv6/fib6_rules.c
index 2f29e4e33bd3..b240f24a6e52 100644
--- a/net/ipv6/fib6_rules.c
+++ b/net/ipv6/fib6_rules.c
@@ -63,19 +63,32 @@ unsigned int fib6_rules_seq_read(struct net *net)
 struct dst_entry *fib6_rule_lookup(struct net *net, struct flowi6 *fl6,
 				   int flags, pol_lookup_t lookup)
 {
-	struct fib_lookup_arg arg = {
-		.lookup_ptr = lookup,
-		.flags = FIB_LOOKUP_NOREF,
-	};
-
-	/* update flow if oif or iif point to device enslaved to l3mdev */
-	l3mdev_update_flow(net, flowi6_to_flowi(fl6));
-
-	fib_rules_lookup(net->ipv6.fib6_rules_ops,
-			 flowi6_to_flowi(fl6), flags, &arg);
-
-	if (arg.result)
-		return arg.result;
+	if (net->ipv6.fib6_has_custom_rules) {
+		struct fib_lookup_arg arg = {
+			.lookup_ptr = lookup,
+			.flags = FIB_LOOKUP_NOREF,
+		};
+
+		/* update flow if oif or iif point to device enslaved to l3mdev */
+		l3mdev_update_flow(net, flowi6_to_flowi(fl6));
+
+		fib_rules_lookup(net->ipv6.fib6_rules_ops,
+				 flowi6_to_flowi(fl6), flags, &arg);
+
+		if (arg.result)
+			return arg.result;
+	} else {
+		struct rt6_info *rt;
+
+		rt = lookup(net, net->ipv6.fib6_local_tbl, fl6, flags);
+		if (rt != net->ipv6.ip6_null_entry && rt->dst.error != -EAGAIN)
+			return &rt->dst;
+		ip6_rt_put(rt);
+		rt = lookup(net, net->ipv6.fib6_main_tbl, fl6, flags);
+		if (rt->dst.error != -EAGAIN)
+			return &rt->dst;
+		ip6_rt_put(rt);
+	}
 
 	dst_hold(&net->ipv6.ip6_null_entry->dst);
 	return &net->ipv6.ip6_null_entry->dst;
@@ -245,6 +258,7 @@ static int fib6_rule_configure(struct fib_rule *rule, struct sk_buff *skb,
 	rule6->dst.plen = frh->dst_len;
 	rule6->tclass = frh->tos;
 
+	net->ipv6.fib6_has_custom_rules = true;
 	err = 0;
 errout:
 	return err;
diff --git a/net/ipv6/route.c b/net/ipv6/route.c
index aba07fce67fb..7ecbe5eb19f8 100644
--- a/net/ipv6/route.c
+++ b/net/ipv6/route.c
@@ -3934,6 +3934,7 @@ static int __net_init ip6_route_net_init(struct net *net)
 			 ip6_template_metrics, true);
 
 #ifdef CONFIG_IPV6_MULTIPLE_TABLES
+	net->ipv6.fib6_has_custom_rules = false;
 	net->ipv6.ip6_prohibit_entry = kmemdup(&ip6_prohibit_entry_template,
 					       sizeof(*net->ipv6.ip6_prohibit_entry),
 					       GFP_KERNEL);
-- 
2.13.3

Re: [PATCH net-next v2] net: ipv6: avoid overhead when no custom FIB rules are installed

From: David Ahern <hidden>
Date: 2017-08-08 19:11:01

On 8/8/17 12:23 PM, Vincent Bernat wrote:
If the user hasn't installed any custom rules, don't go through the
whole FIB rules layer. This is pretty similar to f4530fa574df (ipv4:
Avoid overhead when no custom FIB rules are installed).

Using a micro-benchmark module [1], timing ip6_route_output() with
get_cycles(), with 40,000 routes in the main routing table, before this
patch:
...
At the frequency of the host during the bench (~ 3.7 GHz), this is
about a 100 ns difference on the median value.

A next step would be to collapse local and main tables, as in
0ddcf43d5d4a (ipv4: FIB Local/MAIN table collapse).

[1]: https://github.com/vincentbernat/network-lab/blob/master/lab-routes-ipv6/kbench_mod.c

Signed-off-by: Vincent Bernat <redacted>
Reviewed-by: Jiri Pirko <redacted>
---
 include/net/netns/ipv6.h |  1 +
 net/ipv6/fib6_rules.c    | 40 +++++++++++++++++++++++++++-------------
 net/ipv6/route.c         |  1 +
 3 files changed, 29 insertions(+), 13 deletions(-)
LGTM.

Acked-by: David Ahern <redacted>

Re: [PATCH net-next v2] net: ipv6: avoid overhead when no custom FIB rules are installed

From: David Miller <davem@davemloft.net>
Date: 2017-08-09 04:40:24

From: Vincent Bernat <redacted>
Date: Tue,  8 Aug 2017 20:23:49 +0200
If the user hasn't installed any custom rules, don't go through the
whole FIB rules layer. This is pretty similar to f4530fa574df (ipv4:
Avoid overhead when no custom FIB rules are installed).

Using a micro-benchmark module [1], timing ip6_route_output() with
get_cycles(), with 40,000 routes in the main routing table, before this
patch:

    min=606 max=12911 count=627 average=1959 95th=4903 90th=3747 50th=1602 mad=821
    table=254 avgdepth=21.8 maxdepth=39
    value │                         ┊                            count
      600 │▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒                                         199
      880 │▒▒▒░░░░░░░░░░░░░░░░                                      43
     1160 │▒▒▒░░░░░░░░░░░░░░░░░░░░                                  48
     1440 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░                               43
     1720 │▒▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░                          59
     2000 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                      50
     2280 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                    26
     2560 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                  31
     2840 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░               28
     3120 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              17
     3400 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             17
     3680 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             8
     3960 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           11
     4240 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            6
     4520 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           6
     4800 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           9

After:

    min=544 max=11687 count=627 average=1776 95th=4546 90th=3585 50th=1227 mad=565
    table=254 avgdepth=21.8 maxdepth=39
    value │                         ┊                            count
      540 │▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒                                        201
      800 │▒▒▒▒▒░░░░░░░░░░░░░░░░                                    63
     1060 │▒▒▒▒▒░░░░░░░░░░░░░░░░░░░░░                               68
     1320 │▒▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░                            39
     1580 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                         32
     1840 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                       32
     2100 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                    34
     2360 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░                 33
     2620 │▒▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░               26
     2880 │▒░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              22
     3140 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░              9
     3400 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             8
     3660 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░             9
     3920 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            8
     4180 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           8
     4440 │░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░           8

At the frequency of the host during the bench (~ 3.7 GHz), this is
about a 100 ns difference on the median value.

A next step would be to collapse local and main tables, as in
0ddcf43d5d4a (ipv4: FIB Local/MAIN table collapse).

[1]: https://github.com/vincentbernat/network-lab/blob/master/lab-routes-ipv6/kbench_mod.c

Signed-off-by: Vincent Bernat <redacted>
Reviewed-by: Jiri Pirko <redacted>
Looks great, applied, thanks!
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help