Merge branch 'ipv4-multipath-hash'

Peter Nørlund says:

====================
ipv4: Hash-based multipath routing

When the routing cache was removed in 3.6, the IPv4 multipath algorithm changed
from more or less being destination-based into being quasi-random per-packet
scheduling. This increases the risk of out-of-order packets and makes it
impossible to use multipath together with anycast services.

This patch series replaces the old implementation with flow-based load
balancing based on a hash over the source and destination addresses.

Distribution of the hash is done with thresholds as described in RFC 2992.
This reduces the disruption when a path is added/remove when having more than
two paths.

To futher the chance of successful usage in conjuction with anycast, ICMP
error packets are hashed over the inner IP addresses. This ensures that PMTU
will work together with anycast or load-balancers such as IPVS.

Port numbers are not considered since fragments could cause problems with
anycast and IPVS. Relying on the DF-flag for TCP packets is also insufficient,
since ICMP inspection effectively extracts information from the opposite
flow which might have a different state of the DF-flag. This is also why the
RSS hash is not used. These are typically based on the NDIS RSS spec which
mandates TCP support.

Measurements of the additional overhead of a two-path multipath
(p_mkroute_input excl. __mkroute_input) on a Xeon X3550 (4 cores, 2.66GHz):

Original per-packet: ~394 cycles/packet
L3 hash:              ~76 cycles/packet

Changes in v5:
- Fixed compilation error

Changes in v4:
- Functions take hash directly instead of func ptr
- Added inline hash function
- Added dummy macros to minimize ifdefs
- Use upper 31 bits of hash instead of lower

Changes in v3:
- Multipath algorithm is no longer configurable (always L3)
- Added random seed to hash
- Moved ICMP inspection to isolated function
- Ignore source quench packets (deprecated as per RFC 6633)

Changes in v2:
- Replaced 8-bit xor hash with 31-bit jenkins hash
- Don't scale weights (since 31-bit)
- Avoided unnecesary renaming of variables
- Rely on DF-bit instead of fragment offset when checking for fragmentation
- upper_bound is now inclusive to avoid overflow
- Use a callback to postpone extracting flow information until necessary
- Skipped ICMP inspection entirely with L4 hashing
- Handle newly added sysctl ignore_routes_with_linkdown
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

2 files changed, 21 insertions, 4 deletions

diff --git a/include/net/ip_fib.h b/include/net/ip_fib.h
index 727d6e9a9685..7a51fd8d99e4 100644
--- a/include/net/ip_fib.h
+++ b/include/net/ip_fib.h
@@ -79,7 +79,7 @@ struct fib_nh {
 	unsigned char		nh_scope;
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
 	int			nh_weight;
-	int			nh_power;
+	atomic_t		nh_upper_bound;
 #endif
 #ifdef CONFIG_IP_ROUTE_CLASSID
 	__u32			nh_tclassid;
@@ -118,7 +118,7 @@ struct fib_info {
 #define fib_advmss fib_metrics[RTAX_ADVMSS-1]
 	int			fib_nhs;
 #ifdef CONFIG_IP_ROUTE_MULTIPATH
-	int			fib_power;
+	int			fib_weight;
 #endif
 	struct rcu_head		rcu;
 	struct fib_nh		fib_nh[0];
@@ -320,7 +320,15 @@ int ip_fib_check_default(__be32 gw, struct net_device *dev);
 int fib_sync_down_dev(struct net_device *dev, unsigned long event);
 int fib_sync_down_addr(struct net *net, __be32 local);
 int fib_sync_up(struct net_device *dev, unsigned int nh_flags);
-void fib_select_multipath(struct fib_result *res);
+
+extern u32 fib_multipath_secret __read_mostly;
+
+static inline int fib_multipath_hash(__be32 saddr, __be32 daddr)
+{
+	return jhash_2words(saddr, daddr, fib_multipath_secret) >> 1;
+}
+
+void fib_select_multipath(struct fib_result *res, int hash);
 
 /* Exported by fib_trie.c */
 void fib_trie_init(void);
diff --git a/include/net/route.h b/include/net/route.h
index e211dc167db1..d32cb76f5302 100644
--- a/include/net/route.h
+++ b/include/net/route.h
@@ -28,6 +28,7 @@
 #include <net/inetpeer.h>
 #include <net/flow.h>
 #include <net/inet_sock.h>
+#include <net/ip_fib.h>
 #include <net/l3mdev.h>
 #include <linux/in_route.h>
 #include <linux/rtnetlink.h>
@@ -113,7 +114,15 @@ struct in_device;
 int ip_rt_init(void);
 void rt_cache_flush(struct net *net);
 void rt_flush_dev(struct net_device *dev);
-struct rtable *__ip_route_output_key(struct net *, struct flowi4 *flp);
+struct rtable *__ip_route_output_key_hash(struct net *, struct flowi4 *flp,
+					  int mp_hash);
+
+static inline struct rtable *__ip_route_output_key(struct net *net,
+						   struct flowi4 *flp)
+{
+	return __ip_route_output_key_hash(net, flp, -1);
+}
+
 struct rtable *ip_route_output_flow(struct net *, struct flowi4 *flp,
 				    const struct sock *sk);
 struct dst_entry *ipv4_blackhole_route(struct net *net,