Re: [PATCH] net: add checks if dev->netdev_ops is NULL in register_netdevice()

[Alexander Duyck <alexander.h.duyck@intel.com>]

On 04/08/2013 02:43 AM, dingtianhong wrote:
> In some cases netdev->netdev_ops could be NULL in the register_netdevice(),
> thus a NULL point deference happens and lead to oops:
>
> [ 8260.836400] BUG: unable to handle kernel NULL pointer dereference at (null)
> [ 8260.955306] IP: [<ffffffff81388333>] register_netdevice+0x73/0x2d0
> [ 8261.054888] PGD 2f4dcd067 PUD 2f4de0067 PMD 0
> [ 8261.134244] Oops: 0000 [#1] SMP
> [ 8261.198938] CPU 0
> [ 8262.605182] RIP: 0010:[<ffffffff81388333>]  [<ffffffff81388333>] register_netdevice+0x73/0x2d0
> [ 8262.741859] RSP: 0018:ffff8802f489f9f8  EFLAGS: 00010246
> [ 8262.839463] RAX: 0000000000000000 RBX: ffff8802f4e73000 RCX: ffff8802f4e73006
> [ 8262.959484] RDX: 0000000000000000 RSI: 00000000daa92527 RDI: ffff8802f4e73006
> [ 8263.079352] RBP: 0000000000000000 R08: 0000000000000000 R09: ffff880327d898c0
> [ 8263.198987] R10: 00000000000000c1 R11: 0000000000000000 R12: ffffffff81fed080
> [ 8263.318401] R13: ffffffff81fed080 R14: ffff8802f4e73000 R15: 0000000000000000
> [ 8263.438014] FS:  00007f01dfa81700(0000) GS:ffff88033f200000(0000) knlGS:0000000000000000
> [ 8263.569778] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 8263.673859] CR2: 0000000000000000 CR3: 000000032893f000 CR4: 00000000000006f0
> [ 8263.794922] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [ 8263.915734] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
> [ 8264.036251] Process rmcli (pid: 14089, threadinfo ffff8802f489e000, task ffff880324708080)
> [ 8264.170961] Stack:
> [ 8264.231328]  ffff88032a3c0000 0000000000000003 0000000000000003 ffff88032a3c0000
> [ 8264.356967]  0000000000000003 ffffffffa039dbcb 000000003ec216c0 0000000000000246
> [ 8264.483081]  382e30363238205b 00205d3900000006 ff00305f63696e76 0000000000000003
>
> We should check the point to avoid the oops.
>
> Signed-off-by: Ding Tianhong <dingtianhong@huawei.com>
> ---
>  net/core/dev.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/net/core/dev.c b/net/core/dev.c
> index e7d68ed..596a11e 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -5159,7 +5159,7 @@ int register_netdevice(struct net_device *dev)
>  		goto out;
>
>  	/* Init, if this function is available */
> -	if (dev->netdev_ops->ndo_init) {
> +	if (dev->netdev_ops && dev->netdev_ops->ndo_init) {
>  		ret = dev->netdev_ops->ndo_init(dev);
>  		if (ret) {
>  			if (ret > 0)
> -- 1.8.0
>

The correct fix for this is to not allow the registration of devices
with a null netdev_ops pointer.  The subtle approach to resolving this
is adding a check for the null pointer near the start of the function
and returning -EVINAL since the dev structure passed was configured
incorrectly.  The blunt approach is to just add a BUG_ON with a comment
on the line before it explaining that you should always set the
netdev_ops pointer before trying to register a netdev.  That way if
someone else makes the mistake then they will hit the BUG_ON and realize
what they are doing wrong.

Thanks,

Alex