Re: [PATCH v6 1/4] udp_tunnel: add config option to bind to a device

Le 30 nov. 2018 à 15:31, Sabrina Dubroca [off-list ref] a écrit :

2018-11-27, 14:05:42 +0100, Alexis Bauvin wrote:

quoted

diff --git a/net/ipv4/udp_tunnel.c b/net/ipv4/udp_tunnel.c
index 6539ff15e9a3..dc68e15a4f72 100644
--- a/net/ipv4/udp_tunnel.c
+++ b/net/ipv4/udp_tunnel.c

@@ -20,6 +20,16 @@ int udp_sock_create4(struct net *net, struct udp_port_cfg *cfg,

	if (err < 0)
		goto error;

+	if (cfg->bind_ifindex) {
+		struct net_device *dev;
+
+		dev = __dev_get_by_index(net, cfg->bind_ifindex);

Quoting from net/core/dev.c:

*	[...]                                    The device has not
*	had its reference counter increased so the caller must be careful
*	about locking. The caller must hold either the RTNL semaphore
*	or @dev_base_lock.
*/

which is the case for VXLAN (and GENEVE) during ndo_open, but I don't
think other UDP tunnels (FOU, L2TP) are holding RTNL when they call
udp_sock_create(). dev_get_by_index() + dev_put() should be safe.

Also, I don't think it's a problem with vxlan, but this could handle
the case where __dev_get_by_index returns NULL.

First, sorry for the (late) response, thanks for the feedback!

That’s a real issue I did not take into account when writing the patch; still
too new to kernel dev.

Fixing and spinning up v7.

quoted

+		err = kernel_setsockopt(sock, SOL_SOCKET, SO_BINDTODEVICE,
+					dev->name, strlen(dev->name) + 1);
+		if (err < 0)
+			goto error;
+	}
+
	udp_addr.sin_family = AF_INET;
	udp_addr.sin_addr = cfg->local_ip;
	udp_addr.sin_port = cfg->local_udp_port;

-- 
Sabrina