net: ipv6: Release route when device is unregistering
authorDavid Ahern <dsahern@gmail.com>
Wed, 7 Jun 2017 18:26:23 +0000 (12:26 -0600)
committerDavid S. Miller <davem@davemloft.net>
Thu, 8 Jun 2017 15:12:39 +0000 (11:12 -0400)
Roopa reported attempts to delete a bond device that is referenced in a
multipath route is hanging:

$ ifdown bond2    # ifupdown2 command that deletes virtual devices
unregister_netdevice: waiting for bond2 to become free. Usage count = 2

Steps to reproduce:
    echo 1 > /proc/sys/net/ipv6/conf/all/ignore_routes_with_linkdown
    ip link add dev bond12 type bond
    ip link add dev bond13 type bond
    ip addr add 2001:db8:2::0/64 dev bond12
    ip addr add 2001:db8:3::0/64 dev bond13
    ip route add 2001:db8:33::0/64 nexthop via 2001:db8:2::2 nexthop via 2001:db8:3::2
    ip link del dev bond12
    ip link del dev bond13

The root cause is the recent change to keep routes on a linkdown. Update
the check to detect when the device is unregistering and release the
route for that case.

Fixes: a1a22c12060e4 ("net: ipv6: Keep nexthop of multipath route on admin down")
Reported-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/netdevice.h
net/ipv6/route.c

index ab7ca3fdc495de8e4c88f62d4f81a63f40ed96f3..846193dfb0ac42c1955468a8edc4384dbba6ebbf 100644 (file)
@@ -4262,6 +4262,11 @@ static inline const char *netdev_name(const struct net_device *dev)
        return dev->name;
 }
 
+static inline bool netdev_unregistering(const struct net_device *dev)
+{
+       return dev->reg_state == NETREG_UNREGISTERING;
+}
+
 static inline const char *netdev_reg_state(const struct net_device *dev)
 {
        switch (dev->reg_state) {
index dc61b0b5e64edf7bd69ab905573e38415abf2346..7cebd954d5bb4263017f8f2d577ecfb88ca7c093 100644 (file)
@@ -2804,6 +2804,7 @@ static int fib6_ifdown(struct rt6_info *rt, void *arg)
        if ((rt->dst.dev == dev || !dev) &&
            rt != adn->net->ipv6.ip6_null_entry &&
            (rt->rt6i_nsiblings == 0 ||
+            (dev && netdev_unregistering(dev)) ||
             !rt->rt6i_idev->cnf.ignore_routes_with_linkdown))
                return -1;