diff mbox series

[5.15,19/32] net/mlx5: Bridge, Fix devlink deadlock on net namespace deletion

Message ID 20220204091915.888414403@linuxfoundation.org
State New
Headers show
Series None | expand

Commit Message

Greg KH Feb. 4, 2022, 9:22 a.m. UTC 
From: Roi Dayan <roid@nvidia.com>

commit 880b517691908fb753019b9b27cd082e7617debd upstream.

When changing mode to switchdev, rep bridge init registered to netdevice
notifier holds the devlink lock and then takes pernet_ops_rwsem.
At that time deleting a netns holds pernet_ops_rwsem and then takes
the devlink lock.

Example sequence is:
$ ip netns add foo
$ devlink dev eswitch set pci/0000:00:08.0 mode switchdev &
$ ip netns del foo

deleting netns trace:

[ 1185.365555]  ? devlink_pernet_pre_exit+0x74/0x1c0
[ 1185.368331]  ? mutex_lock_io_nested+0x13f0/0x13f0
[ 1185.370984]  ? xt_find_table+0x40/0x100
[ 1185.373244]  ? __mutex_lock+0x24a/0x15a0
[ 1185.375494]  ? net_generic+0xa0/0x1c0
[ 1185.376844]  ? wait_for_completion_io+0x280/0x280
[ 1185.377767]  ? devlink_pernet_pre_exit+0x74/0x1c0
[ 1185.378686]  devlink_pernet_pre_exit+0x74/0x1c0
[ 1185.379579]  ? devlink_nl_cmd_get_dumpit+0x3a0/0x3a0
[ 1185.380557]  ? xt_find_table+0xda/0x100
[ 1185.381367]  cleanup_net+0x372/0x8e0

changing mode to switchdev trace:

[ 1185.411267]  down_write+0x13a/0x150
[ 1185.412029]  ? down_write_killable+0x180/0x180
[ 1185.413005]  register_netdevice_notifier+0x1e/0x210
[ 1185.414000]  mlx5e_rep_bridge_init+0x181/0x360 [mlx5_core]
[ 1185.415243]  mlx5e_uplink_rep_enable+0x269/0x480 [mlx5_core]
[ 1185.416464]  ? mlx5e_uplink_rep_disable+0x210/0x210 [mlx5_core]
[ 1185.417749]  mlx5e_attach_netdev+0x232/0x400 [mlx5_core]
[ 1185.418906]  mlx5e_netdev_attach_profile+0x15b/0x1e0 [mlx5_core]
[ 1185.420172]  mlx5e_netdev_change_profile+0x15a/0x1d0 [mlx5_core]
[ 1185.421459]  mlx5e_vport_rep_load+0x557/0x780 [mlx5_core]
[ 1185.422624]  ? mlx5e_stats_grp_vport_rep_num_stats+0x10/0x10 [mlx5_core]
[ 1185.424006]  mlx5_esw_offloads_rep_load+0xdb/0x190 [mlx5_core]
[ 1185.425277]  esw_offloads_enable+0xd74/0x14a0 [mlx5_core]

Fix this by registering rep bridges for per net netdev notifier
instead of global one, which operats on the net namespace without holding
the pernet_ops_rwsem.

Fixes: 19e9bfa044f3 ("net/mlx5: Bridge, add offload infrastructure")
Signed-off-by: Roi Dayan <roid@nvidia.com>
Reviewed-by: Vlad Buslov <vladbu@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---
 drivers/net/ethernet/mellanox/mlx5/core/en/rep/bridge.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
diff mbox series

Patch

--- a/drivers/net/ethernet/mellanox/mlx5/core/en/rep/bridge.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en/rep/bridge.c
@@ -491,7 +491,7 @@  void mlx5e_rep_bridge_init(struct mlx5e_
 	}
 
 	br_offloads->netdev_nb.notifier_call = mlx5_esw_bridge_switchdev_port_event;
-	err = register_netdevice_notifier(&br_offloads->netdev_nb);
+	err = register_netdevice_notifier_net(&init_net, &br_offloads->netdev_nb);
 	if (err) {
 		esw_warn(mdev, "Failed to register bridge offloads netdevice notifier (err=%d)\n",
 			 err);
@@ -526,7 +526,7 @@  void mlx5e_rep_bridge_cleanup(struct mlx
 		return;
 
 	cancel_delayed_work_sync(&br_offloads->update_work);
-	unregister_netdevice_notifier(&br_offloads->netdev_nb);
+	unregister_netdevice_notifier_net(&init_net, &br_offloads->netdev_nb);
 	unregister_switchdev_blocking_notifier(&br_offloads->nb_blk);
 	unregister_switchdev_notifier(&br_offloads->nb);
 	destroy_workqueue(br_offloads->wq);