[dpdk-dev] [PATCH] net/bonding: fix socket id check

[Ferruh Yigit]

On 4/27/2021 3:44 AM, Chengchang Tang wrote:
> 
> 
> On 2021/4/26 22:54, Ferruh Yigit wrote:
>> On 4/22/2021 8:12 AM, Min Hu (Connor) wrote:
>>> From: Chengchang Tang <tangchengchang at huawei.com>
>>>
>>> The socket ID entered by user is cast to an unsigned integer. However,
>>> the value may be an illegal negative value, which may cause some
>>> problems. In this case, an error should be returned.
>>>
>>
>> +1 to fix
>>
>>> In addition, the socket ID may be an invalid positive number, which is
>>> also processed in this patch.
>>>
>>> Fixes: 2efb58cbab6e ("bond: new link bonding library")
>>> Cc: stable at dpdk.org
>>>
>>> Signed-off-by: Chengchang Tang <tangchengchang at huawei.com>
>>> Signed-off-by: Min Hu (Connor) <humin29 at huawei.com>
>>> ---
>>>  drivers/net/bonding/rte_eth_bond_args.c | 4 ++--
>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/net/bonding/rte_eth_bond_args.c b/drivers/net/bonding/rte_eth_bond_args.c
>>> index 8c5f90d..bcc0fe3 100644
>>> --- a/drivers/net/bonding/rte_eth_bond_args.c
>>> +++ b/drivers/net/bonding/rte_eth_bond_args.c
>>> @@ -207,12 +207,12 @@ bond_ethdev_parse_socket_id_kvarg(const char *key __rte_unused,
>>>  		return -1;
>>>  
>>>  	errno = 0;
>>> -	socket_id = (uint8_t)strtol(value, &endptr, 10);
>>> +	socket_id = strtol(value, &endptr, 10);
>>
>> 'strtol()' returns 'long int', but implicitly casting it to 'int'. My concern is
>> if this cause a static analysis tool warning.
>> What do you think to have 'socket_id' type as 'long int'?
>>
> I think it would be better to cast to the 'int' here, for reasons below.
> 

Independent from below reasons, converting from user provided "long int" to
'int' will cause losing value and may lead wrong checks,

Like if user provided '-4294967281' (0xffffffff0000000f), when you cast to
'int', it will become '15' (0xf) and will pass from validation checks.

So I think better to verify the value first as "long int", later cast it to 'int'.

>>>  	if (*endptr != 0 || errno != 0)
>>>  		return -1;
>>>  
>>>  	/* validate socket id value */
>>> -	if (socket_id >= 0) {
>>> +	if (socket_id >= 0 && socket_id < RTE_MAX_NUMA_NODES) {>  		*(uint8_t *)extra_args = (uint8_t)socket_id;
>>
>> Here there is an assumption that RTE_MAX_NUMA_NODES will be less than
>> 'UCHAR_MAX', perhaps it can be good to add a check to verify this assumption.
> 
> Currently, it is unlikely that RTE_MAX_NUMA_NODES will be greater than 256. Therefore,
> adding such check will not cause any problems. But I don't think it's necessary to put
> such restrictions on it (i.e. RTE_MAX_NUMA_NODES should be less than UCHAR_MAX).

Restriction comes from provided 'extra_args' being 'uint8_t', I just suggest
checking this.

> I checked all references to RTE_MAX_NUMA_NODES, and usually socket_id is of type 'int'
> or 'unsigned int' (Only the efd, node, and bonding specify 'unsigned char' for socket
> IDs.). And for that, I think it will be better to change the socket id type to 'int'
> in this patch. For the type of socket id in efd and node, I will send new patches to
> modify it.
>>
>>>  		return 0;
>>>  	}
>>>
>>
>>
>> .
>>
>