[DTrace-devel] [PATCH] test: Use copyinstr() and copyin() when using data in user buffers

[Eugene Loh]

On 2/25/23 23:05, Kris Van Hees wrote:

> On Sat, Feb 25, 2023 at 03:19:06PM -0500, eugene.loh--- via DTrace-devel wrote:
>> diff --git a/test/unittest/funcs/strjoin/tst.strjoin_nonDPTR.d b/test/unittest/funcs/strjoin/tst.strjoin_nonDPTR.d
>> @@ -17,7 +17,7 @@ BEGIN
>>   syscall::write:entry
>>   /ppid == $pid/
>>   {
>> -	printf("|%s|\n", strjoin((void *)arg1, (void *)arg1));
>> +	printf("|%s|\n", strjoin(copyinstr(arg1), copyinstr(arg1)));
> The problem with this change is that you go from pointers that you cannot
> access directly on some platforms to using copyinstr() which copies the
> strings into tstrings before calling strjoin() which (I think) invalidates
> the testing.  Sure, tstrings are not marked DPTR but they are in DTrace
> managed memory anyway.  In other words, if the strjoin() were to invalidly
> use indirect loads on the pointers it would actually work.
>
> Instead, I think we should use something like:
>
> 	printf("|%s|\n", strjoin(`linux_banner, `linux_banner));
>
> And I would expect the same applies for the other tests...
>
> And those that use copyin() will end up with the data in ALLOCA memory which
> again allows indirect loads...  So that is an issue also.

Hmm.  Yes I suppose so.  Okay, I'll replace this patch with something 
entitled

     test: Use kernel pointer for nonDPTR tests