ipv4: Fallback to FIB local table in __ip_dev_find().

In commit 9435eb1
("ipv4: Implement __ip_dev_find using new interface address hash.")
we reimplemented __ip_dev_find() so that it doesn't have to
do a full FIB table lookup.

Instead, it consults a hash table of addresses configured to
interfaces.

This works identically to the old code in all except one case,
and that is for loopback subnets.

The old code would match the loopback device for any IP address
that falls within a subnet configured to the loopback device.

Handle this corner case by doing the FIB lookup.

We could implement this via inet_addr_onlink() but:

1) Someone could configure many addresses to loopback and
   inet_addr_onlink() is a simple list traversal.

2) We know the old code works.

Reported-by: Julian Anastasov <ja@ssi.bg>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/devinet.c

@@ -64,6 +64,8 @@
 #include <net/rtnetlink.h>
 #include <net/net_namespace.h>
 
+#include "fib_lookup.h"
+
 static struct ipv4_devconf ipv4_devconf = {
 	.data = {
 		[IPV4_DEVCONF_ACCEPT_REDIRECTS - 1] = 1,
@@ -151,6 +153,20 @@ struct net_device *__ip_dev_find(struct net *net, __be32 addr, bool devref)
 			break;
 		}
 	}
+	if (!result) {
+		struct flowi4 fl4 = { .daddr = addr };
+		struct fib_result res = { 0 };
+		struct fib_table *local;
+
+		/* Fallback to FIB local table so that communication
+		 * over loopback subnets work.
+		 */
+		local = fib_get_table(net, RT_TABLE_LOCAL);
+		if (local &&
+		    !fib_table_lookup(local, &fl4, &res, FIB_LOOKUP_NOREF) &&
+		    res.type == RTN_LOCAL)
+			result = FIB_RES_DEV(res);
+	}
 	if (result && devref)
 		dev_hold(result);
 	rcu_read_unlock();