From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ralf Baechle Subject: Re: instant oops with AF_ROSE Date: Wed, 20 Jul 2011 10:32:14 +0100 Message-ID: <20110720093214.GA20152@linux-mips.org> References: <4E186E25.5080200@free.fr> Mime-Version: 1.0 Return-path: Content-Disposition: inline In-Reply-To: <4E186E25.5080200@free.fr> Sender: linux-hams-owner@vger.kernel.org List-ID: Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Bernard F6BVP Cc: wchast@gmail.com, davem@davemloft.net, linux-hams@vger.kernel.org On Sat, Jul 09, 2011 at 05:05:09PM +0200, Bernard F6BVP wrote: > What we do with ROSE nodes is to create a network allowing AX25 > frames routing based on destination address. However, basically, > ROSE frames > are still AX.25 standard frames that can be sent over packet radio. > In order to relay rose frames through Internet, applications like > ax25ipd encapsulates rose inside of IP or UDP frames. > > I agree that rose module is still unfriendly and will crash linux > kernel on some occasions. > For example, after creating rose socket and populating rose routing > table, removing rose module is not possible. > > There is either a kernel crash or a looping message > > kernel:unregister_netdevice: waiting for rose0 to become free. Usage > count = 6 > > where count can be any number without any relation to actual usage > and remaining the same, preventing Linux to shutdown. > > I noticed that there are no refcount entries in rose node or rose neigh > structures. This may be related ? As part of fixing the problem reported by David I've found 4 race conditions so far. I've posted the patches to linux-hams for testing but on a 2nd thought I think the race accessing rose_neigh->use is unlikely to be the cause for what Bernard is observing. Ralf