Re: [PATCH] rdmavt: avoid NULL pointer dereference in rvt_qp_exit()

[Dennis Dalessandro]

On 10/18/22 4:41 AM, Leon Romanovsky wrote:
> On Mon, Oct 17, 2022 at 05:26:52PM +0300, Natalia Petrova wrote:
>> rvt_qp_exit() checks 'rdi->qp_dev' for NULL, but the pointer is
>> dereferenced before that in rvt_free_all_qps().
>>
>> Found by Linux Verification Center (linuxtesting.org) with SVACE.
>>
>> Fixes: f92e48718889 ("IB/rdmavt: Reset all QPs when the device is shut
>> down")
> 
> Please never break fixes line.
> 
>> Signed-off-by: Natalia Petrova <n.petrova@xxxxxxxxxx>
>> Signed-off-by: Alexey Khoroshilov <khoroshilov@xxxxxxxxx>
>> ---
>>  drivers/infiniband/sw/rdmavt/qp.c | 9 ++++++---
>>  1 file changed, 6 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/infiniband/sw/rdmavt/qp.c b/drivers/infiniband/sw/rdmavt/qp.c
>> index 3acab569fbb9..06e755975f61 100644
>> --- a/drivers/infiniband/sw/rdmavt/qp.c
>> +++ b/drivers/infiniband/sw/rdmavt/qp.c
>> @@ -459,13 +459,16 @@ static unsigned rvt_free_all_qps(struct rvt_dev_info *rdi)
>>   */
>>  void rvt_qp_exit(struct rvt_dev_info *rdi)
>>  {
>> -	u32 qps_inuse = rvt_free_all_qps(rdi);
>> +	u32 qps_inuse = 0;
>> +
>> +	if (!rdi->qp_dev)
>> +		return;
>> +
>> +	qps_inuse = rvt_free_all_qps(rdi);
> 
> These lines are not needed.
> 
>>  
>>  	if (qps_inuse)
>>  		rvt_pr_err(rdi, "QP memory leak! %u still in use\n",
>>  			   qps_inuse);
>> -	if (!rdi->qp_dev)
>> -		return;
> 
> It is enough to delete these two lines. At this stage, rdi->qp_dev always
> exists as it was created in rvt_register_device().
> 

Agree with Leon here. qp_dev is created in rvt_register_device which will fail
if the qp dev allocation fails in rvt_driver_qp_init().

-Denny