The test that nla_put_iflink() uses to detect whether a device has a
lower link doesn't work with network namespaces, as a device can have
the same ifindex as its parent:

    ip netns add main
    ip netns add peer
    ip -net main link add dummy0 type dummy
    ip -net main link add link dummy0 macvlan0 netns peer type macvlan
    ip -net main link show type dummy
        # 9: dummy0: <BROADCAST,NOARP> mtu 1500 qdisc noop ...
    ip -net peer link show type macvlan
        # 9: macvlan0@if9: <BROADCAST,MULTICAST> mtu 1500 qdisc noop ...

Instead of calling dev_get_iflink(), we can use the existence of the
ndo_get_iflink operation (which dev_get_iflink would call) to check if
a device has a lower link.

I previously tried to fix this with commit feadc4b6cf42 ("rtnetlink:
always put IFLA_LINK for links with a link-netnsid") but didn't get to
the root of the problem.

Fixes: d8a5ec672768 ("[NET]: netlink support for moving devices between network 
namespaces.")
Signed-off-by: Sabrina Dubroca <s...@queasysnail.net>
---
 net/core/rtnetlink.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index c35b3f02b4f9..a8459fb59ccd 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -1549,12 +1549,13 @@ static int put_master_ifindex(struct sk_buff *skb, 
struct net_device *dev)
 
 static int nla_put_iflink(struct sk_buff *skb, const struct net_device *dev)
 {
-       int ifindex = dev_get_iflink(dev);
+       if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) {
+               int ifindex = dev->netdev_ops->ndo_get_iflink(dev);
 
-       if (dev->ifindex == ifindex)
-               return 0;
+               return nla_put_u32(skb, IFLA_LINK, ifindex);
+       }
 
-       return nla_put_u32(skb, IFLA_LINK, ifindex);
+       return 0;
 }
 
 static noinline_for_stack int nla_put_ifalias(struct sk_buff *skb,
-- 
2.28.0

Reply via email to