Message-ID: <CAJieiUhFsA799SjSVaYPtpEajH5rAD907ObPXoZ4KYKjvxMeTg@mail.gmail.com>
Date:   Mon, 26 Nov 2018 21:58:03 -0800
From:   Roopa Prabhu <roopa@...ulusnetworks.com>
To:     abauvin@...leway.com
Cc:     David Ahern <dsa@...ulusnetworks.com>,
        Stephen Hemminger <stephen@...workplumber.org>,
        netdev <netdev@...r.kernel.org>, akherbouche@...leway.com
Subject: Re: [PATCH v5 5/6] vxlan: handle underlay VRF changes

On Mon, Nov 26, 2018 at 5:04 PM Alexis Bauvin <abauvin@...leway.com> wrote:
>
> When underlay VRF changes, either because the lower device itself changed,
> or its VRF changed, this patch releases the current socket of the VXLAN
> device and recreates another one in the right VRF. This allows for
> on-the-fly change of the underlay VRF of a VXLAN device.
>
> Signed-off-by: Alexis Bauvin <abauvin@...leway.com>
> Reviewed-by: Amine Kherbouche <akherbouche@...leway.com>
> Tested-by: Amine Kherbouche <akherbouche@...leway.com>
> ---

re-iterating my comments on the patch this time.

this version still unconditionally calls reopen even if the current
state of the device is closed (eg vxlan_stop).
generally not in favor of the unconditional open/close in the driver.
Lets see if there are other options.
I interpreted one of Davids suggestions to force the change ordering
from user-space by returning an error.
ie Make the user do a down and up of the vxlan device if he wants to
change the vrf of the default remote dev.

This patch needs more thought, the rest are ok to go in if you
separate them out.

>  drivers/net/vxlan.c | 82 +++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 82 insertions(+)
>
> diff --git a/drivers/net/vxlan.c b/drivers/net/vxlan.c
> index 8ba0a57ff958..131ee80a38f9 100644
> --- a/drivers/net/vxlan.c
> +++ b/drivers/net/vxlan.c
> @@ -3720,6 +3720,33 @@ struct net_device *vxlan_dev_create(struct net *net, const char *name,
>  }
>  EXPORT_SYMBOL_GPL(vxlan_dev_create);
>
> +static int vxlan_reopen(struct vxlan_net *vn, struct vxlan_dev *vxlan)
> +{
> +       int ret = 0;
> +
> +       if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
> +           !vxlan_group_used(vn, vxlan))
> +               ret = vxlan_igmp_leave(vxlan);
> +       vxlan_sock_release(vxlan);
> +
> +       if (ret < 0)
> +               return ret;
> +
> +       ret = vxlan_sock_add(vxlan);
> +       if (ret < 0)
> +               return ret;
> +
> +       if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
> +               ret = vxlan_igmp_join(vxlan);
> +               if (ret == -EADDRINUSE)
> +                       ret = 0;
> +               if (ret)
> +                       vxlan_sock_release(vxlan);
> +       }
> +
> +       return ret;
> +}
> +
>  static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
>                                              struct net_device *dev)
>  {
> @@ -3742,6 +3769,55 @@ static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
>         unregister_netdevice_many(&list_kill);
>  }
>
> +static void vxlan_handle_change_upper(struct vxlan_net *vn,
> +                                     struct net_device *dev)
> +{
> +       struct vxlan_dev *vxlan, *next;
> +
> +       list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
> +               struct net_device *lower;
> +               int err;
> +
> +               lower = __dev_get_by_index(vxlan->net,
> +                                          vxlan->cfg.remote_ifindex);
> +               if (!netdev_is_upper_master(lower, dev))
> +                       continue;
> +
> +               err = vxlan_reopen(vn, vxlan);
> +               if (err < 0)
> +                       netdev_err(vxlan->dev, "Failed to reopen socket: %d\n",
> +                                  err);
> +       }
> +}
> +
> +static void vxlan_handle_change(struct vxlan_net *vn, struct net_device *dev)
> +{
> +       struct vxlan_dev *vxlan = netdev_priv(dev);
> +       struct vxlan_sock *sock;
> +       int l3mdev_index = 0;
> +
> +#if IS_ENABLED(CONFIG_IPV6)
> +       bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
> +       bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
> +
> +       sock = ipv6 ? rcu_dereference(vxlan->vn6_sock)
> +                   : rcu_dereference(vxlan->vn4_sock);
> +#else
> +       sock = rcu_dereference(vxlan->vn4_sock);
> +#endif
> +
> +       if (vxlan->cfg.remote_ifindex)
> +               l3mdev_index = l3mdev_master_upper_ifindex_by_index(
> +                       vxlan->net, vxlan->cfg.remote_ifindex);
> +       if (sock->sock->sk->sk_bound_dev_if != l3mdev_index) {
> +               int ret = vxlan_reopen(vn, vxlan);
> +
> +               if (ret < 0)
> +                       netdev_err(vxlan->dev, "Failed to reopen socket: %d\n",
> +                                  ret);
> +       }
> +}
> +
>  static int vxlan_netdevice_event(struct notifier_block *unused,
>                                  unsigned long event, void *ptr)
>  {
> @@ -3756,6 +3832,12 @@ static int vxlan_netdevice_event(struct notifier_block *unused,
>         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
>                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
>                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
> +       } else if (event == NETDEV_CHANGEUPPER) {
> +               vxlan_handle_change_upper(vn, dev);
> +       } else if (event == NETDEV_CHANGE) {
> +               if (dev->rtnl_link_ops &&
> +                   !strcmp(dev->rtnl_link_ops->kind, vxlan_link_ops.kind))
> +                       vxlan_handle_change(vn, dev);

This should move to the rtnl changelink handler


>         }
>
>         return NOTIFY_DONE;
> --
>

Powered by blists - more mailing lists