Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Fri, 13 Dec 2019 18:10:51 -0800
From:   Jakub Kicinski <jakub.kicinski@netronome.com>
To:     Matteo Croce <mcroce@redhat.com>
Cc:     netdev@vger.kernel.org, Jay Vosburgh <j.vosburgh@gmail.com>,
        Veaceslav Falico <vfalico@gmail.com>,
        Andy Gospodarek <andy@greyhouse.net>,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH net-next] bonding: don't init workqueues on error
Message-ID: <20191213181051.0f949b17@cakuba.netronome.com>
In-Reply-To: <20191210152454.86247-1-mcroce@redhat.com>
References: <20191210152454.86247-1-mcroce@redhat.com>
Organization: Netronome Systems, Ltd.
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, 10 Dec 2019 16:24:54 +0100, Matteo Croce wrote:
> bond_create() initialize six workqueues used later on.

Work _entries_ not _queues_ no?

> In the unlikely event that the device registration fails, these
> structures are initialized unnecessarily, so move the initialization
> out of the error path. Also, create an error label to remove some
> duplicated code.

Does the initialization of work entries matter? Is this prep for further
changes?

> Signed-off-by: Matteo Croce <mcroce@redhat.com>
> ---
>  drivers/net/bonding/bond_main.c | 11 +++++++----
>  1 file changed, 7 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
> index fcb7c2f7f001..8756b6a023d7 100644
> --- a/drivers/net/bonding/bond_main.c
> +++ b/drivers/net/bonding/bond_main.c
> @@ -4889,8 +4889,8 @@ int bond_create(struct net *net, const char *name)
>  				   bond_setup, tx_queues);
>  	if (!bond_dev) {
>  		pr_err("%s: eek! can't alloc netdev!\n", name);

If this is a clean up patch I think this pr_err() could also be removed?
Memory allocation usually fail very loudly so there should be no reason
to print more errors.

> -		rtnl_unlock();
> -		return -ENOMEM;
> +		res = -ENOMEM;
> +		goto out_unlock;
>  	}
>  
>  	/*
> @@ -4905,14 +4905,17 @@ int bond_create(struct net *net, const char *name)
>  	bond_dev->rtnl_link_ops = &bond_link_ops;
>  
>  	res = register_netdevice(bond_dev);
> +	if (res < 0) {
> +		free_netdev(bond_dev);
> +		goto out_unlock;
> +	}
>  
>  	netif_carrier_off(bond_dev);
>  
>  	bond_work_init_all(bond);
>  
> +out_unlock:
>  	rtnl_unlock();
> -	if (res < 0)
> -		free_netdev(bond_dev);
>  	return res;
>  }
>  

I do appreciate that the change makes the error handling follow a more
usual kernel pattern, but IMHO it'd be even better if the error
handling was completely moved. IOW the success path should end with
return 0; and the error path should contain free_netdev(bond_dev);

-	int res;
+	int err;

	[...]

	rtnl_unlock();

	return 0;

err_free_netdev:
	free_netdev(bond_dev);
err_unlock:
	rtnl_unlock();
	return err;

I'm just not 100% sold on the improvement made by this patch being
worth the code churn, please convince me, respin or get an ack from 
one of the maintainers? :)