Re: [PATCH net] net: rtnl_register in net_ns_init need rtnl_lock

On 22.09.2016 15:03, Eric Dumazet wrote:

On Thu, 2016-09-22 at 13:03 +0200, Hannes Frederic Sowa wrote:

quoted

Signed-off-by: Hannes Frederic Sowa <redacted>
---
 net/core/net_namespace.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/net/core/net_namespace.c b/net/core/net_namespace.c
index 2c2eb1b629b11d..a2ace299f28355 100644
--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c

@@ -758,9 +758,11 @@ static int __init net_ns_init(void)
 
 	register_pernet_subsys(&net_ns_ops);
 
+	rtnl_lock();
 	rtnl_register(PF_UNSPEC, RTM_NEWNSID, rtnl_net_newid, NULL, NULL);
 	rtnl_register(PF_UNSPEC, RTM_GETNSID, rtnl_net_getid, rtnl_net_dumpid,
 		      NULL);
+	rtnl_unlock();
 
 	return 0;
 }

Hi Hannes

Why is this needed here, and not in other places ?

I found this during working on the file and actually saw no live issues
(belonged to another series which I just split up).

I don't think it is a big issue but wanted the writes to the
rtnl_msg_handlers array to be strictly serialized. I was working on
adding this to other places, too. Maybe better for net-next even?

Theoretically we would need to add a memory barriers to make sure we
don't publish uninitialized memory into the array if concurrent readers
of the array want to find their function pointers.

Hint : A changelog always help reviewers and future bug hunting.

I will add that to v2.

Thanks,
Hannes