syscalls/brk: add direct syscall tst_variant

Direct usage of brk is discouraged in favor of using malloc. Also, brk was
removed from POSIX in POSIX.1-2001.
In particular, the Musl libc's brk always returns -ENOMEM which causes
the LTP tests to exit prematurely. Invoking the syscall directly allows
them to properly validate brk behavior. Add a new test variant handling if
the libc wrappers are not implemented and testing the direct syscall.

Use tst_syscall() and handle the failure cases ourselves, as
we don't depend on the libc to do it anymore.

The patch also works around the dependency on sbrk to get the current break
as it has the same issues. Instead, call tst_syscall(__NR_brk, 0) which
always returns the current break.

Update brk01 to use void* to unify it with brk02.

Signed-off-by: Teo Couprie Diaz <teo.coupriediaz@arm.com>
---
v1: Reworked from RFC by adding a variant rather than replacing the
    existing tests. Thanks Petr and Cyril !

Not sure this is the best way to do it. Maybe it could use a shared header
for testing if the libc version is implemented ?

Green CI: https://github.com/Teo-CD/ltp/actions/runs/3629342833/jobs/6121433817

 testcases/kernel/syscalls/brk/brk01.c | 36 +++++++++++++++++------
 testcases/kernel/syscalls/brk/brk02.c | 42 ++++++++++++++++++++++-----
 2 files changed, 62 insertions(+), 16 deletions(-)


base-commit: 990c0b48f8508a747863b1dea5556fba57e1c304

Hi Teo,

> Direct usage of brk is discouraged in favor of using malloc. Also, brk was
> removed from POSIX in POSIX.1-2001.
> In particular, the Musl libc's brk always returns -ENOMEM which causes
> the LTP tests to exit prematurely. Invoking the syscall directly allows
> them to properly validate brk behavior. Add a new test variant handling if
> the libc wrappers are not implemented and testing the direct syscall.

> Use tst_syscall() and handle the failure cases ourselves, as
> we don't depend on the libc to do it anymore.

> The patch also works around the dependency on sbrk to get the current break
> as it has the same issues. Instead, call tst_syscall(__NR_brk, 0) which
> always returns the current break.

> Update brk01 to use void* to unify it with brk02.
That brought warnings, see below.

...
> +++ b/testcases/kernel/syscalls/brk/brk02.c
> @@ -14,24 +14,51 @@
>  #include <unistd.h>
>  #include <sys/mman.h>
>  #include "tst_test.h"
> +#include "lapi/syscalls.h"
> +
> +inline void *brk_variants(void *addr)
> +{
> +	void *brk_addr;
> +	if (tst_variant) {
> +		brk_addr = (void *)tst_syscall(__NR_brk, addr);
> +	} else {
> +		brk_addr = (void *)brk(addr);

NOTE using pointer for brk() is problematic, there are complains:

brk02.c: In function ‘brk_variants’:
brk02.c:26:28: warning: cast to pointer from integer of different size
[-Wint-to-pointer-cast]
   26 |                 brk_addr = (void *)brk(addr);
         |                            ^

Not sure how to fix this. Problem is in both C files, on several places.

There are also code style problems (you can see it when running make check in
testcases/kernel/syscalls/brk/), but that's a minor detail.

Kind regards,
Petr

Hi Petr,

On 06/12/2022 14:04, Petr Vorel wrote:
> Hi Teo,
>
>> Direct usage of brk is discouraged in favor of using malloc. Also, brk was
>> removed from POSIX in POSIX.1-2001.
>> In particular, the Musl libc's brk always returns -ENOMEM which causes
>> the LTP tests to exit prematurely. Invoking the syscall directly allows
>> them to properly validate brk behavior. Add a new test variant handling if
>> the libc wrappers are not implemented and testing the direct syscall.
>> Use tst_syscall() and handle the failure cases ourselves, as
>> we don't depend on the libc to do it anymore.
>> The patch also works around the dependency on sbrk to get the current break
>> as it has the same issues. Instead, call tst_syscall(__NR_brk, 0) which
>> always returns the current break.
>> Update brk01 to use void* to unify it with brk02.
> That brought warnings, see below.
>
> ...
>> +++ b/testcases/kernel/syscalls/brk/brk02.c
>> @@ -14,24 +14,51 @@
>>   #include <unistd.h>
>>   #include <sys/mman.h>
>>   #include "tst_test.h"
>> +#include "lapi/syscalls.h"
>> +
>> +inline void *brk_variants(void *addr)
>> +{
>> +	void *brk_addr;
>> +	if (tst_variant) {
>> +		brk_addr = (void *)tst_syscall(__NR_brk, addr);
>> +	} else {
>> +		brk_addr = (void *)brk(addr);
> NOTE using pointer for brk() is problematic, there are complains:
>
> brk02.c: In function ‘brk_variants’:
> brk02.c:26:28: warning: cast to pointer from integer of different size
> [-Wint-to-pointer-cast]
>     26 |                 brk_addr = (void *)brk(addr);
>           |                            ^
>
> Not sure how to fix this. Problem is in both C files, on several places.

Ah yes, of course. That's my mistake: I just realized that I used the 
libc brk as if it returned the break, but it doesn't. It just returns an 
error, so the warning is justified (and a drowned in other messages from 
my build system).

I'll rework and change that, I believe it will work without warnings 
once it is properly used.

> There are also code style problems (you can see it when running make check in
> testcases/kernel/syscalls/brk/), but that's a minor detail.
I wasn't aware of that, good to know thanks ! I will take care of it as 
well.
> Kind regards,
> Petr
Thanks for giving it a look,
Téo