From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Garzik Subject: Re: [PATCH] fix tuntap oversight Date: Mon, 12 Apr 2004 12:43:30 -0400 Sender: linux-kernel-owner@vger.kernel.org Message-ID: <407AC732.1090000@pobox.com> References: <20040412065947.GC18810@net-ronin.org> <20040412001551.05476658.davem@redhat.com> <20040412162916.GA5046@net-ronin.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Cc: "David S. Miller" , netdev@oss.sgi.com, linux-kernel@vger.kernel.org Return-path: To: Jeremy Martin In-Reply-To: <20040412162916.GA5046@net-ronin.org> List-Id: netdev.vger.kernel.org Jeremy Martin wrote: > On Mon, Apr 12, 2004 at 12:15:51AM -0700, David S. Miller wrote: > >>This netif_running() check is not necessary, and in fact >>wrong. >> >>In fact, if ethernet drivers erroneously do this, this causes >>them to fail to support the ALB bonding driver modes which >>require on-the-fly MAC address changes while the interface is >>up. >> > > > I just took a look in drivers/net/ > and > acenic.c > atarilance.c > b44.c > cs89x0.c > net_init.c > typhoon.c > > all use that netif_running() check when setting the MAC. I actually just pulled > the function from net_init.c for the tun change. Are these broken? > (I'm asking in total ignorance so be gentle :). It's different for a driver that drives real hardware. struct net_device::set_mac_address() is called inside rtnl_lock(). The safe thing to do is 1) read MAC address from eeprom on probe 2) write MAC address to hardware upon each dev->open() 3) use default eth_mac_addr() from net_init.c And the netif_running() check in eth_mac_addr() is correct, because it does not update the hardware MAC address (which in this API would be impossible). Normally the netif_running() check is for hardware that cannot update its MAC address safely during operation. Jeff