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

[Chengchang Tang]

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.

>>  	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).
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;
>>  	}
>>
> 
> 
> .
>