syscalls/ioctl_ns01: fix crash on aarch64

Test crashes with SIGBUS when using child stack. Align stack to 256 bytes,
which is more than enough for any arch.

Neither parent or library is waiting for child process. Add SIGCHLD to
clone flags.

Check return value of ltp_clone(), and TBROK on failure.

Fix warning about unused *arg.

Signed-off-by: Jan Stancek <jstancek@redhat.com>
---
 testcases/kernel/syscalls/ioctl/ioctl_ns01.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

On Tue, Jun 11, 2019 at 3:44 PM Jan Stancek <jstancek@redhat.com> wrote:

> Test crashes with SIGBUS when using child stack. Align stack to 256 bytes,
> which is more than enough for any arch.


> Neither parent or library is waiting for child process. Add SIGCHLD to
> clone flags.
>
> Check return value of ltp_clone(), and TBROK on failure.
>
> Fix warning about unused *arg.
>
> Signed-off-by: Jan Stancek <jstancek@redhat.com>
> ---
>  testcases/kernel/syscalls/ioctl/ioctl_ns01.c | 10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
>
> diff --git a/testcases/kernel/syscalls/ioctl/ioctl_ns01.c
> b/testcases/kernel/syscalls/ioctl/ioctl_ns01.c
> index dfde4da6c5d6..625de9bd832d 100644
> --- a/testcases/kernel/syscalls/ioctl/ioctl_ns01.c
> +++ b/testcases/kernel/syscalls/ioctl/ioctl_ns01.c
> @@ -23,7 +23,7 @@
>
>  #define STACK_SIZE (1024 * 1024)
>
> -static char child_stack[STACK_SIZE];
> +static char child_stack[STACK_SIZE] __attribute__((aligned(256)));
>

This patch makes sense. And maybe we'd better change that for
ioctl_nfs05/06 too?

BTW, another way we could try is to allocate the child_stack memory
dynamically via malloc(STACK_SIZE) in setup() function.


>
>  static void setup(void)
>  {
> @@ -53,7 +53,7 @@ static void test_ns_get_parent(void)
>         }
>  }
>
> -static int child(void *arg)
> +static int child(void *arg LTP_ATTRIBUTE_UNUSED)
>  {
>         test_ns_get_parent();
>         return 0;
> @@ -61,10 +61,14 @@ static int child(void *arg)
>
>  static void run(void)
>  {
> +       int child_pid;
> +
>         test_ns_get_parent();
>
> -       ltp_clone(CLONE_NEWPID, &child, 0,
> +       child_pid = ltp_clone(CLONE_NEWPID | SIGCHLD, &child, 0,
>                 STACK_SIZE, child_stack);
> +       if (child_pid == -1)
> +               tst_brk(TBROK | TERRNO, "ltp_clone failed");
>  }
>
>  static struct tst_test test = {
> --
> 1.8.3.1
>
>
> --
> Mailing list info: https://lists.linux.it/listinfo/ltp
>

----- Original Message -----

> On Tue, Jun 11, 2019 at 3:44 PM Jan Stancek < jstancek@redhat.com > wrote:

> > Test crashes with SIGBUS when using child stack. Align stack to 256 bytes,
> 
> > which is more than enough for any arch.
> 
> > Neither parent or library is waiting for child process. Add SIGCHLD to
> 
> > clone flags.
> 

> > Check return value of ltp_clone(), and TBROK on failure.
> 

> > Fix warning about unused *arg.
> 

> > Signed-off-by: Jan Stancek < jstancek@redhat.com >
> 
> > ---
> 
> > testcases/kernel/syscalls/ioctl/ioctl_ns01.c | 10 +++++++---
> 
> > 1 file changed, 7 insertions(+), 3 deletions(-)
> 

> > diff --git a/testcases/kernel/syscalls/ioctl/ioctl_ns01.c
> > b/testcases/kernel/syscalls/ioctl/ioctl_ns01.c
> 
> > index dfde4da6c5d6..625de9bd832d 100644
> 
> > --- a/testcases/kernel/syscalls/ioctl/ioctl_ns01.c
> 
> > +++ b/testcases/kernel/syscalls/ioctl/ioctl_ns01.c
> 
> > @@ -23,7 +23,7 @@
> 

> > #define STACK_SIZE (1024 * 1024)
> 

> > -static char child_stack[STACK_SIZE];
> 
> > +static char child_stack[STACK_SIZE] __attribute__((aligned(256)));
> 

> This patch makes sense. And maybe we'd better change that for ioctl_nfs05/06
> too?

Yes, you're right. Trying an unaligned address yields same result: 

Starting program: /root/ltp/testcases/kernel/syscalls/ioctl/ioctl_ns05 
tst_test.c:1111: INFO: Timeout per run is 0h 05m 00s 
[Attaching after process 19449 fork to child process 19452] 
[New inferior 2 (process 19452)] 
[Detaching after fork from parent process 19449] 
[Inferior 1 (process 19449) detached] 
0x4306a0 
[New LWP 19453] 
ioctl_ns05.c:80: PASS: child and parent are consistent 

Thread 2.2 received signal SIGBUS, Bus error. 
[Switching to LWP 19453] 
child (arg=0x0) at ioctl_ns05.c:37 
37 if (getpid() != 1) 
(gdb) disassemble 
Dump of assembler code for function child: 
=> 0x00000000004032e8 <+0>: stp x29, x30, [sp, #-32]! 
0x00000000004032ec <+4>: mov x29, sp 
0x00000000004032f0 <+8>: str x19, [sp, #16] 

I'll send v2. 

Thanks, 
Jan 

> BTW, another way we could try is to allocate the child_stack memory
> dynamically via malloc(STACK_SIZE) in setup() function.

> > static void setup(void)
> 
> > {
> 
> > @@ -53,7 +53,7 @@ static void test_ns_get_parent(void)
> 
> > }
> 
> > }
> 

> > -static int child(void *arg)
> 
> > +static int child(void *arg LTP_ATTRIBUTE_UNUSED)
> 
> > {
> 
> > test_ns_get_parent();
> 
> > return 0;
> 
> > @@ -61,10 +61,14 @@ static int child(void *arg)
> 

> > static void run(void)
> 
> > {
> 
> > + int child_pid;
> 
> > +
> 
> > test_ns_get_parent();
> 

> > - ltp_clone(CLONE_NEWPID, &child, 0,
> 
> > + child_pid = ltp_clone(CLONE_NEWPID | SIGCHLD, &child, 0,
> 
> > STACK_SIZE, child_stack);
> 
> > + if (child_pid == -1)
> 
> > + tst_brk(TBROK | TERRNO, "ltp_clone failed");
> 
> > }
> 

> > static struct tst_test test = {
> 
> > --
> 
> > 1.8.3.1
> 

> > --
> 
> > Mailing list info: https://lists.linux.it/listinfo/ltp
> 

> --
> Regards,
> Li Wang