leds: trigger: netdev: fix RTNL handling to prevent potential deadlock

When working on LED support for r8169 I got the following lockdep
warning. Easiest way to prevent this scenario seems to be to take
the RTNL lock before the trigger_data lock in set_device_name().

======================================================
WARNING: possible circular locking dependency detected
6.7.0-rc2-next-20231124+ #2 Not tainted
------------------------------------------------------
bash/383 is trying to acquire lock:
ffff888103aa1c68 (&trigger_data->lock){+.+.}-{3:3}, at: netdev_trig_notify+0xec/0x190 [ledtrig_netdev]

but task is already holding lock:
ffffffff8cddf808 (rtnl_mutex){+.+.}-{3:3}, at: rtnl_lock+0x12/0x20

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (rtnl_mutex){+.+.}-{3:3}:
       __mutex_lock+0x9b/0xb50
       mutex_lock_nested+0x16/0x20
       rtnl_lock+0x12/0x20
       set_device_name+0xa9/0x120 [ledtrig_netdev]
       netdev_trig_activate+0x1a1/0x230 [ledtrig_netdev]
       led_trigger_set+0x172/0x2c0
       led_trigger_write+0xf1/0x140
       sysfs_kf_bin_write+0x5d/0x80
       kernfs_fop_write_iter+0x15d/0x210
       vfs_write+0x1f0/0x510
       ksys_write+0x6c/0xf0
       __x64_sys_write+0x14/0x20
       do_syscall_64+0x3f/0xf0
       entry_SYSCALL_64_after_hwframe+0x6c/0x74

-> #0 (&trigger_data->lock){+.+.}-{3:3}:
       __lock_acquire+0x1459/0x25a0
       lock_acquire+0xc8/0x2d0
       __mutex_lock+0x9b/0xb50
       mutex_lock_nested+0x16/0x20
       netdev_trig_notify+0xec/0x190 [ledtrig_netdev]
       call_netdevice_register_net_notifiers+0x5a/0x100
       register_netdevice_notifier+0x85/0x120
       netdev_trig_activate+0x1d4/0x230 [ledtrig_netdev]
       led_trigger_set+0x172/0x2c0
       led_trigger_write+0xf1/0x140
       sysfs_kf_bin_write+0x5d/0x80
       kernfs_fop_write_iter+0x15d/0x210
       vfs_write+0x1f0/0x510
       ksys_write+0x6c/0xf0
       __x64_sys_write+0x14/0x20
       do_syscall_64+0x3f/0xf0
       entry_SYSCALL_64_after_hwframe+0x6c/0x74

other info that might help us debug this:

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(rtnl_mutex);
                               lock(&trigger_data->lock);
                               lock(rtnl_mutex);
  lock(&trigger_data->lock);

 *** DEADLOCK ***

8 locks held by bash/383:
 #0: ffff888103ff33f0 (sb_writers#3){.+.+}-{0:0}, at: ksys_write+0x6c/0xf0
 #1: ffff888103aa1e88 (&of->mutex){+.+.}-{3:3}, at: kernfs_fop_write_iter+0x114/0x210
 #2: ffff8881036f1890 (kn->active#82){.+.+}-{0:0}, at: kernfs_fop_write_iter+0x11d/0x210
 #3: ffff888108e2c358 (&led_cdev->led_access){+.+.}-{3:3}, at: led_trigger_write+0x30/0x140
 #4: ffffffff8cdd9e10 (triggers_list_lock){++++}-{3:3}, at: led_trigger_write+0x75/0x140
 #5: ffff888108e2c270 (&led_cdev->trigger_lock){++++}-{3:3}, at: led_trigger_write+0xe3/0x140
 torvalds#6: ffffffff8cdde3d0 (pernet_ops_rwsem){++++}-{3:3}, at: register_netdevice_notifier+0x1c/0x120
 torvalds#7: ffffffff8cddf808 (rtnl_mutex){+.+.}-{3:3}, at: rtnl_lock+0x12/0x20

stack backtrace:
CPU: 0 PID: 383 Comm: bash Not tainted 6.7.0-rc2-next-20231124+ #2
Hardware name: Default string Default string/Default string, BIOS ADLN.M6.SODIMM.ZB.CY.015 08/08/2023
Call Trace:
 <TASK>
 dump_stack_lvl+0x5c/0xd0
 dump_stack+0x10/0x20
 print_circular_bug+0x2dd/0x410
 check_noncircular+0x131/0x150
 __lock_acquire+0x1459/0x25a0
 lock_acquire+0xc8/0x2d0
 ? netdev_trig_notify+0xec/0x190 [ledtrig_netdev]
 __mutex_lock+0x9b/0xb50
 ? netdev_trig_notify+0xec/0x190 [ledtrig_netdev]
 ? __this_cpu_preempt_check+0x13/0x20
 ? netdev_trig_notify+0xec/0x190 [ledtrig_netdev]
 ? __cancel_work_timer+0x11c/0x1b0
 ? __mutex_lock+0x123/0xb50
 mutex_lock_nested+0x16/0x20
 ? mutex_lock_nested+0x16/0x20
 netdev_trig_notify+0xec/0x190 [ledtrig_netdev]
 call_netdevice_register_net_notifiers+0x5a/0x100
 register_netdevice_notifier+0x85/0x120
 netdev_trig_activate+0x1d4/0x230 [ledtrig_netdev]
 led_trigger_set+0x172/0x2c0
 ? preempt_count_add+0x49/0xc0
 led_trigger_write+0xf1/0x140
 sysfs_kf_bin_write+0x5d/0x80
 kernfs_fop_write_iter+0x15d/0x210
 vfs_write+0x1f0/0x510
 ksys_write+0x6c/0xf0
 __x64_sys_write+0x14/0x20
 do_syscall_64+0x3f/0xf0
 entry_SYSCALL_64_after_hwframe+0x6c/0x74
RIP: 0033:0x7f269055d034
Code: c7 00 16 00 00 00 b8 ff ff ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 f3 0f 1e fa 80 3d 35 c3 0d 00 00 74 13 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 54 c3 0f 1f 00 48 83 ec 28 48 89 54 24 18 48
RSP: 002b:00007ffddb7ef748 EFLAGS: 00000202 ORIG_RAX: 0000000000000001
RAX: ffffffffffffffda RBX: 0000000000000007 RCX: 00007f269055d034
RDX: 0000000000000007 RSI: 000055bf5f4af3c0 RDI: 0000000000000001
RBP: 000055bf5f4af3c0 R08: 0000000000000073 R09: 0000000000000001
R10: 0000000000000000 R11: 0000000000000202 R12: 0000000000000007
R13: 00007f26906325c0 R14: 00007f269062ff20 R15: 0000000000000000
 </TASK>

Fixes: d5e0126 ("leds: trigger: netdev: add additional specific link speed mode")
Cc: [email protected]
Signed-off-by: Heiner Kallweit <[email protected]>
Reviewed-by: Andrew Lunn <[email protected]>
Acked-by: Lee Jones <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

drivers/leds/trigger/ledtrig-netdev.c

@@ -226,6 +226,11 @@ static int set_device_name(struct led_netdev_data *trigger_data,
 
 	cancel_delayed_work_sync(&trigger_data->work);
 
+	/*
+	 * Take RTNL lock before trigger_data lock to prevent potential
+	 * deadlock with netdev notifier registration.
+	 */
+	rtnl_lock();
 	mutex_lock(&trigger_data->lock);
 
 	if (trigger_data->net_dev) {
@@ -245,16 +250,14 @@ static int set_device_name(struct led_netdev_data *trigger_data,
 	trigger_data->carrier_link_up = false;
 	trigger_data->link_speed = SPEED_UNKNOWN;
 	trigger_data->duplex = DUPLEX_UNKNOWN;
-	if (trigger_data->net_dev != NULL) {
-		rtnl_lock();
+	if (trigger_data->net_dev)
 		get_device_state(trigger_data);
-		rtnl_unlock();
-	}
 
 	trigger_data->last_activity = 0;
 
 	set_baseline_state(trigger_data);
 	mutex_unlock(&trigger_data->lock);
+	rtnl_unlock();
 
 	return 0;
 }