Resent again, last attempt still altered the plain text.


Zhengchao Shao <shaozhengchao@huawei.com> identified by inspection bugs in the error path of hdlcdev_open() in synclink_gt.c

The function did not fully unwind actions in the error path. The use of try_module_get()/module_put() is unnecessary, potentially hazardous and is removed. The synclink_gt driver is already pinned any point the net device is registered, a requirement for calling this entry point.

The call hdlc_open() to init the generic HDLC layer is moved to after driver level init/checks and proper rollback of previous actions is added. This is a more sensible ordering as the most common error paths are at the driver level and the driver level rollbacks require less processing than hdlc_open()/hdlc_close().

This has been tested with supported hardware.

Signed-off-by:Paul Fulghum <paulkf@microgate.com>

diff --git a/drivers/tty/synclink_gt.c b/drivers/tty/synclink_gt.c
index 25e9befdda3a..72b76cdde534 100644
--- a/drivers/tty/synclink_gt.c
+++ b/drivers/tty/synclink_gt.c
@@ -1433,16 +1433,8 @@ static int hdlcdev_open(struct net_device *dev)
 	int rc;
 	unsigned long flags;
 
-	if (!try_module_get(THIS_MODULE))
-		return -EBUSY;
-
 	DBGINFO(("%s hdlcdev_open\n", dev->name));
 
-	/* generic HDLC layer open processing */
-	rc = hdlc_open(dev);
-	if (rc)
-		return rc;
-
 	/* arbitrate between network and tty opens */
 	spin_lock_irqsave(&info->netlock, flags);
 	if (info->port.count != 0 || info->netcount != 0) {
@@ -1461,6 +1453,16 @@ static int hdlcdev_open(struct net_device *dev)
 		return rc;
 	}
 
+	/* generic HDLC layer open processing */
+	rc = hdlc_open(dev);
+	if (rc) {
+		shutdown(info);
+		spin_lock_irqsave(&info->netlock, flags);
+		info->netcount = 0;
+		spin_unlock_irqrestore(&info->netlock, flags);
+		return rc;
+	}
+
 	/* assert RTS and DTR, apply hardware settings */
 	info->signals |= SerialSignal_RTS | SerialSignal_DTR;
 	program_hw(info);
@@ -1506,7 +1508,6 @@ static int hdlcdev_close(struct net_device *dev)
 	info->netcount=0;
 	spin_unlock_irqrestore(&info->netlock, flags);
 
-	module_put(THIS_MODULE);
 	return 0;
 }
 

On Tue, Nov 15, 2022 at 09:38:32AM -0800, Paul Fulghum wrote:
> Resent again, last attempt still altered the plain text.
> 

Please send a v2 patch, and this is not needed as it would show up in
the commit log if we were to apply it, right?

> 
> Zhengchao Shao <shaozhengchao@huawei.com> identified by inspection bugs in the error path of hdlcdev_open() in synclink_gt.c

Properly wrap your lines at 72 columns please.

> 
> The function did not fully unwind actions in the error path. The use of try_module_get()/module_put() is unnecessary, potentially hazardous and is removed. The synclink_gt driver is already pinned any point the net device is registered, a requirement for calling this entry point.
> 
> The call hdlc_open() to init the generic HDLC layer is moved to after driver level init/checks and proper rollback of previous actions is added. This is a more sensible ordering as the most common error paths are at the driver level and the driver level rollbacks require less processing than hdlc_open()/hdlc_close().
> 
> This has been tested with supported hardware.
> 
> Signed-off-by:Paul Fulghum <paulkf@microgate.com>

You need a Suggested-by: tag here.

And a space after the ':' character.

> 
> diff --git a/drivers/tty/synclink_gt.c b/drivers/tty/synclink_gt.c
> index 25e9befdda3a..72b76cdde534 100644
> --- a/drivers/tty/synclink_gt.c
> +++ b/drivers/tty/synclink_gt.c
> @@ -1433,16 +1433,8 @@ static int hdlcdev_open(struct net_device *dev)
>  	int rc;
>  	unsigned long flags;
>  
> -	if (!try_module_get(THIS_MODULE))
> -		return -EBUSY;

Thank you for removing this, this code pattern is always wrong :)

thanks,

greg k-h