Steven Rostedt wrote:
quoted text
> On Fri, 7 Aug 2009, Li Zefan wrote:
> 
>>>>  	if (is_string_field(field)) {
>>>> +		pred->str_len = field->size;
>>>> +
>>>>  		if (field->filter_type == FILTER_STATIC_STRING)
>>>>  			fn = filter_pred_string;
>>>> -		else
>>>> +		else if (field->filter_type == FILTER_DYN_STRING)
>>>>  			fn = filter_pred_strloc;
>>>> -		pred->str_len = field->size;
>>>> +		else {
>>>> +			fn = filter_pred_pchar;
>>>> +			pred->str_len = strlen(pred->str_val);
>>>> +		}
>>> I'm a little dense here, where do we protect against someone making a 
>>> tracepoint that points to unsafe data?
>>>
>> We can't prevent anyone from doing insane things deliberately, but
>> we prevent from doing wrong things unconsciously.
>>
>> Only if a TRACE_EVENT has a field defined as:
>>
>> 	__field_ext(char *, name, FILTER_PTR_STR)
>>
>> Here using FILTER_PTR_STR explicitly, he should know what he's doing.
>>
>> Anyway, he can make a ptr pointing to unsafe data this way:
>>
>> 	TP_STRUCT__entry(
>> 		__field(char *, name)
>> 	)
>> 	TP_printk("%s", name)
> 
> I guess the thing I'm missing is what's the difference of the two? Why 
> would a developer use __field_ext instead of doing it the unsafe way of 
> just __field?
> 
> I guess I don't see the developer doing something wrong unconsciously. 
> Well maybe I don't see this making the developer do it right 
> unconsciously.
> 
> What protection is this giving us?
> 

__field(char *) suggests it should be treated as plain pointer,
while __field_ext(char *, FILTER_PTR_STR) suggests he's aware it's
safe to dereference the pointer, for example the case in Frederic's
blk events.

In Frederic's initial version, "char *" field will always be
attached to ptr_str filter function. This is unsafe, because for
other fields defined as "char *" but not safe to dereference,
a user still can do this:

	# echo 'name == abc' > filter

Then we'll deref a pointer that can point to unsafe data.

In this patch, this won't happen, as long as the developer is
aware that his use of __field_ext(char *) is right.

Otherwise, he will just use normal __field(char *) and print
the pointer itself in TP_printk().

--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/