ipv4: fix refcount leak in fib_check_nh()
authorEric Dumazet <edumazet@google.com>
Sat, 15 Aug 2015 17:54:07 +0000 (10:54 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 17 Aug 2015 05:14:32 +0000 (22:14 -0700)
fib_lookup() forces FIB_LOOKUP_NOREF flag, while fib_table_lookup()
does not.

This patch solves the typical message at reboot time or device
dismantle :

unregister_netdevice: waiting for eth0 to become free. Usage count = 4

Fixes: 3bfd847203c6 ("net: Use passed in table for nexthop lookups")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: David Ahern <dsa@cumulusnetworks.com>
Acked-by: David Ahern <dsa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/fib_semantics.c

index b7f1d20a96154b0295ece902def96d637622affe..c8025851dac720bd2075c57a9ef64e042fe415c4 100644 (file)
@@ -708,7 +708,8 @@ static int fib_check_nh(struct fib_config *cfg, struct fib_info *fi,
 
                        if (tbl)
                                err = fib_table_lookup(tbl, &fl4, &res,
-                                                  FIB_LOOKUP_IGNORE_LINKSTATE);
+                                                      FIB_LOOKUP_IGNORE_LINKSTATE |
+                                                      FIB_LOOKUP_NOREF);
                        else
                                err = fib_lookup(net, &fl4, &res,
                                                 FIB_LOOKUP_IGNORE_LINKSTATE);