Re: [PATCH v5] Introduce support for Systems Management Driver over WMI for Dell Systems

[Hans de Goede <hdegoede@redhat.com>]

HI,

On 10/6/20 10:46 AM, Bharathi, Divya wrote:
> Sorry for another mail on same patch. I missed to add one response on
> previous comments

No problem.

> <snip>
> 
>>> +/**
>>> + * init_bios_attributes() - Initialize all attributes for a type
>>> + * @attr_type: The attribute type to initialize
>>> + * @guid: The WMI GUID associated with this type to initialize
>>> + *
>>> + * Initialiaze all 4 types of attributes enumeration, integer, string and
>> password object.
>>> + * Populates each attrbute typ's respective properties under sysfs files
>>> + **/
>>> +static int init_bios_attributes(int attr_type, const char *guid)
>>> +{
>>> +	struct kobject *attr_name_kobj; //individual attribute names
>>> +	union acpi_object *obj = NULL;
>>> +	union acpi_object *elements;
>>> +	struct kset *tmp_set;
>>> +
>>> +	/* instance_id needs to be reset for each type GUID
>>> +	 * also, instance IDs are unique within GUID but not across
>>> +	 */
>>> +	int instance_id = 0;
>>> +	int retval = 0;
>>> +
>>> +	retval = alloc_attributes_data(attr_type);
>>> +	if (retval)
>>> +		return retval;
>>> +	/* need to use specific instance_id and guid combination to get right
>> data */
>>> +	obj = get_wmiobj_pointer(instance_id, guid);
>>> +	if (!obj)
>>> +		return -ENODEV;
>>> +	elements = obj->package.elements;
>>> +
>>> +	mutex_lock(&wmi_priv.mutex);
>>> +	while (elements) {
>>> +		/* sanity checking */
>>> +		if (strlen(elements[ATTR_NAME].string.pointer) == 0) {
>>> +			pr_debug("empty attribute found\n");
>>> +			goto nextobj;
>>> +		}
>>> +		if (attr_type == PO)
>>> +			tmp_set = wmi_priv.authentication_dir_kset;
>>> +		else
>>> +			tmp_set = wmi_priv.main_dir_kset;
>>> +
>>> +		if (kset_find_obj(tmp_set,
>> elements[ATTR_NAME].string.pointer)) {
>>> +			pr_debug("duplicate attribute name found - %s\n",
>>> +				elements[ATTR_NAME].string.pointer);
>>> +			goto nextobj;
>>> +		}
>>> +
>>> +		/* build attribute */
>>> +		attr_name_kobj = kzalloc(sizeof(*attr_name_kobj),
>> GFP_KERNEL);
>>> +		if (!attr_name_kobj)
>>> +			goto err_attr_init;
>>> +
>>> +		attr_name_kobj->kset = tmp_set;
>>> +
>>> +		retval = kobject_init_and_add(attr_name_kobj,
>> &attr_name_ktype, NULL, "%s",
>>> +
>> 	elements[ATTR_NAME].string.pointer);
>>> +		if (retval) {
>>> +			kobject_put(attr_name_kobj);
>>> +			goto err_attr_init;
>>> +		}
>>> +
>>> +		/* enumerate all of this attribute */
>>> +		switch (attr_type) {
>>> +		case ENUM:
>>> +			retval = populate_enum_data(elements, instance_id,
>> attr_name_kobj);
>>> +			break;
>>> +		case INT:
>>> +			retval = populate_int_data(elements, instance_id,
>> attr_name_kobj);
>>> +			break;
>>> +		case STR:
>>> +			retval = populate_str_data(elements, instance_id,
>> attr_name_kobj);
>>> +			break;
>>> +		case PO:
>>> +			retval = populate_po_data(elements, instance_id,
>> attr_name_kobj);
>>> +			break;
>>> +		default:
>>> +			break;
>>> +		}
>>
>> The show functions don't take the mutex and can be called as soon as
>> kobject_init_and_add() is called, so it would be better to first populate the
>> data for the current instance_id and only then call kobject_init_and_add()
>>
> 
> Populate functions called here for each type of attribute uses
> attribute_koject which helps in attribute group cleanup.

Good point, and we do allocate the data before creating the kobjects,
so if a user manages to hit the race (which almost certainly would have
to be done intentionally) then the read would just result in an empty
string (rather then say a null pointer dereference oops).

So lets just keep this as is.

Regards,

Hans