On Wed, May 13, 2015 at 09:38:41PM -0700, Eric Dumazet wrote:

And only one of the socket is listed twice (ffff881f6eceb000)

Apparently this is the one _after_ kernel socket.

Does it ring a bell ?

No that's just a walker bug.  In netlink_seq_next when we come
around the loop after ++i, we fail to refresh the ht pointer so
we walk the previous table again.

This bug was introduced in 3.18:

commit 78fd1d0ab072d4d9b5f0b7c14a1516665170b565
Author: Thomas Graf [off-list ref]
Date:   Tue Oct 21 22:05:38 2014 +0200

    netlink: Re-add locking to netlink_lookup() and seq walker

This should have zero impact on functionality though.

Cheers,
-- 
Email: Herbert Xu [off-list ref]
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt