Re: [PATCH net] net: rtnetlink: prevent underflows in do_setvfinfo()

[Parav Pandit <parav@mellanox.com>]

On 11/21/2019 4:44 PM, David Miller wrote:
> From: Dan Carpenter <dan.carpenter@oracle.com>
> Date: Wed, 20 Nov 2019 15:34:38 +0300
>
>> The "ivm->vf" variable is a u32, but the problem is that a number of
>> drivers cast it to an int and then forget to check for negatives.  An
>> example of this is in the cxgb4 driver.
>>
>> drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c
>>   2890  static int cxgb4_mgmt_get_vf_config(struct net_device *dev,
>>   2891                                      int vf, struct ifla_vf_info *ivi)
>>                                             ^^^^^^
>>   2892  {
>>   2893          struct port_info *pi = netdev_priv(dev);
>>   2894          struct adapter *adap = pi->adapter;
>>   2895          struct vf_info *vfinfo;
>>   2896  
>>   2897          if (vf >= adap->num_vfs)
>>                     ^^^^^^^^^^^^^^^^^^^
>>   2898                  return -EINVAL;
>>   2899          vfinfo = &adap->vfinfo[vf];
>>                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
>>
>> There are 48 functions affected.
>  ...
>> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> I'm going to apply this and queue it up for -stable.
>
> The u32 conversion should happen in next.
okay.
When Dan reported, I reviewed and wrote the patch for u32, and some more
code consolidation with it.
But hit the block with u32 leading to overflow to 0 which modifies the
first VF incorrectly.
Checking num_vfs of pci dev, without device_lock() is equally buggy as
it stands today.
Creating a lock in each vendor driver(s) doesn't make sense.
Haven't had chance yet after that.
Let me know if someone has a suggestion that I should incorporate.