hashbase-mpath-select-patch

fib_semantics.patch

*** linux-source-3.13.0-original/net/ipv4/fib_semantics.c	Tue Dec 16 09:23:29 2014
--- linux-source-3.13.0/net/ipv4/fib_semantics.c	Wed Jan 21 03:02:27 2015
***************
*** 33,38 ****
--- 33,42 ----
  #include <linux/init.h>
  #include <linux/slab.h>
  
+ #ifdef CONFIG_IP_ROUTE_MULTIPATH_HASHONLY
+ #include <linux/hash.h>	/* hash_32() */
+ #endif
+ 
  #include <net/arp.h>
  #include <net/ip.h>
  #include <net/protocol.h>
*************** void fib_select_multipath(struct fib_res
*** 1329,1332 ****
--- 1333,1365 ----
  	res->nh_sel = 0;
  	spin_unlock_bh(&fib_multipath_lock);
  }
+ 
+ #endif
+ 
+ #ifdef CONFIG_IP_ROUTE_MULTIPATH_HASHONLY
+ /* added by [email protected]
+  * If hash-only decides next hop for a flow by hash of 5 tuple. there
+  * is no weight based balancing (like commodity hardware routers).
+  */
+ 
+ static inline unsigned long fib_calculate_hash (struct flowi4 * fl4)
+ {
+ 	return hash_32 (fl4->saddr + fl4->daddr + fl4->flowi4_proto + 
+ 			fl4->fl4_sport + fl4->fl4_dport, 8);
+ }
+ 
+ void
+ fib_select_multipath_hashonly(struct fib_result *res,
+ 			      const struct flowi4 * fl4)
+ {
+ 	unsigned long key;
+ 	struct fib_info *fi = res->fi;
+ 
+ 	key = fib_calculate_hash (fl4);
+ 	res->nh_sel = key % fi->fib_nhs;
+ 
+ 	return;
+ }
+ 
  #endif
+ 

ip_output.patch

*** linux-source-3.13.0-original/net/ipv4/ip_output.c	Tue Dec 16 09:23:29 2014
--- linux-source-3.13.0/net/ipv4/ip_output.c	Wed Jan 21 03:48:14 2015
*************** int ip_queue_xmit(struct sk_buff *skb, s
*** 335,340 ****
--- 335,350 ----
  	if (rt != NULL)
  		goto packet_routed;
  
+ #ifdef CONFIG_IP_ROUTE_MULTIPATH_HASHONLY
+ 	iph = ip_hdr (skb);
+         if (iph->protocol == IPPROTO_TCP) {
+ 		fl4->fl4_sport = tcp_hdr (skb)->source;
+ 		fl4->fl4_dport = tcp_hdr (skb)->dest;
+ 	} else if (iph->protocol == IPPROTO_UDP) {
+ 		fl4->fl4_sport = udp_hdr (skb)->source;
+ 		fl4->fl4_dport = udp_hdr (skb)->dest;
+ 	}
+ #endif
  	/* Make sure we can route this packet. */
  	rt = (struct rtable *)__sk_dst_check(sk, 0);
  	if (rt == NULL) {

Kconfig.patch

*** linux-source-3.13.0-original/net/ipv4/Kconfig	Mon Jan 20 11:40:07 2014
--- linux-source-3.13.0/net/ipv4/Kconfig	Wed Jan 21 01:05:44 2015
*************** config IP_ROUTE_MULTIPATH
*** 90,95 ****
--- 90,106 ----
  	  equal "cost" and chooses one of them in a non-deterministic fashion
  	  if a matching packet arrives.
  
+ config IP_ROUTE_MULTIPATH_HASHONLY
+ 	bool "IP: equal cost multipath without load balancing for flows"
+ 	depends on IP_ROUTE_MULTIPATH
+ 	help
+ 	  normally, IP_ROUTE_MULTIPATH load balance single flow to multiple
+ 	  next hops. However, it causes packet reodering so that commodity
+ 	  hardware routers using only hash to decide next hop for individual
+ 	  flows. This option emulates this behavior. A next hop for a flow
+ 	  is decided by hash value for 5-tuple of the flow, and the next hop
+ 	  have been never changed.
+ 
  config IP_ROUTE_VERBOSE
  	bool "IP: verbose route monitoring"
  	depends on IP_ADVANCED_ROUTER

route.patch

*** linux-source-3.13.0-original/net/ipv4/route.c	Tue Dec 16 09:23:29 2014
--- linux-source-3.13.0/net/ipv4/route.c	Wed Jan 21 04:03:29 2015
*************** static int ip_mkroute_input(struct sk_bu
*** 1632,1645 ****
  			    __be32 daddr, __be32 saddr, u32 tos)
  {
  #ifdef CONFIG_IP_ROUTE_MULTIPATH
! 	if (res->fi && res->fi->fib_nhs > 1)
  		fib_select_multipath(res);
  #endif
  
  	/* create a routing cache entry */
  	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
  }
  
  /*
   *	NOTE. We drop all the packets that has local source
   *	addresses, because every properly looped back packet
--- 1632,1667 ----
  			    __be32 daddr, __be32 saddr, u32 tos)
  {
  #ifdef CONFIG_IP_ROUTE_MULTIPATH
! 	if (res->fi && res->fi->fib_nhs > 1) {
! #ifndef CONFIG_IP_ROUTE_MULTIPATH_HASHONLY
  		fib_select_multipath(res);
+ #else
+ 		fib_select_multipath_hashonly(res, fl4);
  #endif
+ #endif
+ 	}
  
  	/* create a routing cache entry */
  	return __mkroute_input(skb, res, in_dev, daddr, saddr, tos);
  }
  
+ #ifdef CONFIG_IP_ROUTE_MULTIPATH_HASHONLY
+ static inline void
+ ip_rt_build_fl4_port (struct sk_buff *skb, struct flowi4 * fl4)
+ {
+ 	struct iphdr *ip = (struct iphdr *) skb_network_header (skb);
+ 
+ 	if (ip->protocol == IPPROTO_TCP) {
+ 		fl4->fl4_sport = tcp_hdr (skb)->source;
+ 		fl4->fl4_dport = tcp_hdr (skb)->dest;
+ 	} else if (ip->protocol == IPPROTO_UDP) {
+ 		fl4->fl4_sport = udp_hdr (skb)->source;
+ 		fl4->fl4_dport = udp_hdr (skb)->dest;
+ 	}
+ 	return;
+ }
+ #endif
+ 
  /*
   *	NOTE. We drop all the packets that has local source
   *	addresses, because every properly looped back packet
*************** static int ip_route_input_slow(struct sk
*** 1710,1715 ****
--- 1732,1740 ----
  	fl4.flowi4_scope = RT_SCOPE_UNIVERSE;
  	fl4.daddr = daddr;
  	fl4.saddr = saddr;
+ #ifdef CONFIG_IP_ROUTE_MULTIPATH_HASHONLY
+ 	ip_rt_build_fl4_port (skb, &fl4);
+ #endif
  	err = fib_lookup(net, &fl4, &res);
  	if (err != 0)
  		goto no_route;
*************** struct rtable *__ip_route_output_key(str
*** 2159,2167 ****
  	}
  
  #ifdef CONFIG_IP_ROUTE_MULTIPATH
! 	if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0)
  		fib_select_multipath(&res);
! 	else
  #endif
  	if (!res.prefixlen &&
  	    res.table->tb_num_default > 1 &&
--- 2184,2196 ----
  	}
  
  #ifdef CONFIG_IP_ROUTE_MULTIPATH
! 	if (res.fi->fib_nhs > 1 && fl4->flowi4_oif == 0) {
! #ifndef CONFIG_IP_ROUTE_MULTIPATH_HASHONLY
  		fib_select_multipath(&res);
! #else
! 		fib_select_multipath_hashonly(&res, fl4);
! #endif
! 	} else
  #endif
  	if (!res.prefixlen &&
  	    res.table->tb_num_default > 1 &&