[PATCH] Fix netpoll cleanup on abort without dev

[PATCH] Fix netpoll cleanup on abort without dev

[Matt Mackall]

If netpoll attempts to use a device without polling support, it will
oops when shutting down. This adds a check that we've actually
attached to a device.

Signed-off-by: Matt Mackall <mpm@selenic.com>

Index: np/net/core/netpoll.c
===================================================================
--- np.orig/net/core/netpoll.c	2004-08-04 08:50:56.000000000 -0500
+++ np/net/core/netpoll.c	2004-08-04 09:02:50.000000000 -0500
@@ -614,7 +614,8 @@
 		spin_lock_irqsave(&rx_list_lock, flags);
 		list_del(&np->rx_list);
 #ifdef CONFIG_NETPOLL_RX
-		np->dev->netpoll_rx = 0;
+		if (np->dev)
+			np->dev->netpoll_rx = 0;
 #endif
 		spin_unlock_irqrestore(&rx_list_lock, flags);
 	}


-- 
Mathematics is the supreme nostalgia of our time.

Re: [PATCH] Fix netpoll cleanup on abort without dev

[Jeff Moyer]

==> Regarding [PATCH] Fix netpoll cleanup on abort without dev; Matt Mackall <mpm@selenic.com> adds:

mpm> If netpoll attempts to use a device without polling support, it will
mpm> oops when shutting down. This adds a check that we've actually
mpm> attached to a device.

Hi, Matt,

Perhaps I'm missing something, but how do we successfully return from
netpoll_setup without np->dev filled in?  Netpoll_setup has the following:

        if (!ndev->poll_controller) {
                printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
                       np->name, np->dev_name);
                goto release;
        }


Thanks,

Jeff

Re: [PATCH] Fix netpoll cleanup on abort without dev

[Matt Mackall]

On Wed, Aug 04, 2004 at 10:54:01AM -0400, Jeff Moyer wrote:
> ==> Regarding [PATCH] Fix netpoll cleanup on abort without dev; Matt Mackall <mpm@selenic.com> adds:
> 
> mpm> If netpoll attempts to use a device without polling support, it will
> mpm> oops when shutting down. This adds a check that we've actually
> mpm> attached to a device.
> 
> Hi, Matt,
> 
> Perhaps I'm missing something, but how do we successfully return from
> netpoll_setup without np->dev filled in?  Netpoll_setup has the following:
> 
>         if (!ndev->poll_controller) {
>                 printk(KERN_ERR "%s: %s doesn't support polling, aborting.\n",
>                        np->name, np->dev_name);
>                 goto release;
>         }

We don't. What happens is netconsole is blindly calling
netpoll_cleanup on module exit. I could move the np->dev check up to
that point, but it seems to me that netpoll_cleanup should be made
harmless instead.

ps: you still owe me a netpoll patch

-- 
Mathematics is the supreme nostalgia of our time.