[apparmor] [patch] tests: readdir - test both getdents() and getdents64() if available

[Tyler Hicks]

On 04/04/2017 03:24 PM, Steve Beattie wrote:
> Hey Tyler,
> 
> On Tue, Apr 04, 2017 at 02:03:53PM -0500, Tyler Hicks wrote:
>> On 04/04/2017 01:14 PM, Steve Beattie wrote:
>>> -int main(int argc, char *argv[])
>>> +#ifdef SYS_getdents
>>> +int my_readdir(char *dirname)
>>
>> Nitpick...
>>
>> Please use my_getdents() and my_getdents64() since these are getdents
>> wrappers and not readdir wrappers.
> 
> The reason for calling the function my_readdir() and not my_getdents()
> is the same reason the whole test is called the same; they're emulating
> the work that the libc readdir() function does. while ensuring that
> glibc doesn't do anything underneath us that might invalidate our
> assumptions about the test. But I'm not too chuffed one or the other
> about the name.

I didn't even think about the name of the test. :)

Feel free to keep it my_readdir().

> 
>>> +	rc = my_readdir(argv[1]);
>>> +	if (rc != 0)
>>> +		goto err;
>>> +
>>> +	rc = my_readdir64(argv[1]);
>>> +	if (rc != 0)
>>> +		goto err;
>>
>> Objection...
>>
>> In an expected fail test case, getdents64() will not be tested because
>> getdents() will have already failed.
>>
>> The test script should pass the expected behavior to the test program
>> and the test program will need to verify that both syscalls match the
>> expected behavior.
>>
>> Alternatively, a shortcut would be to test both syscalls and if the
>> results are different, raise an ERROR instead of a FAIL.
> 
> Yes, I thought about this a bit before submitting. The tricky situation
> is cases like aarch64 (aka arm64) where the getdents() syscall doesn't
> exist at all and the stub function is used instead. The test would
> need to take this into account, perhaps by having the stub function
> return -ENOSYS or something, to make the situation detectable.
> 
> I'll think about this more.

Riiight. Maybe it is easier to stick an ifdef or two in the calling
function. If both are defined, expect equality in the return code. If
only one is defined, don't do an return code equality check.

I didn't mean to make this simple test improvement turn into something
complex. I'm willing to ack your original patch if you don't see a quick
and easy solution to this issue.

Tyler

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: OpenPGP digital signature
URL: <https://lists.ubuntu.com/archives/apparmor/attachments/20170404/ff14a254/attachment-0001.pgp>