[klibc] a little help w/ sys_clone

[H. Peter Anvin]

christophe varoqui wrote:
> 
> How does it impact the correctness of the following snippet on IA64 ?
> 

sys_clone doesn't exist, you'd need to use sys_clone2.  In general you 
probably should use the libc wrapper: the _syscall*() macros are broken on 
about half of all platforms.  IIRC sys_clone2 takes a stack size as well as a 
stack pointer.

> #include <sched.h>
> #include <linux/unistd.h>
> 
> #define __NR_sys_clone __NR_clone
> 
> static inline _syscall2(int,sys_clone, int,flag, void*,stack);
> 
> int main(void)
> {
>         int pid;
> 
>         pid = sys_clone(CLONE_NEWNS, 0);
>         if (pid < 0)
>                 exit(1);
>         if (pid > 0)
>                 exit(0);
>         /* in child */
>         return 0;
> }
> 
> 
>>Note: klibc is not multithreaded.
> 
> Yes, not that annoying : only the multipah daemon is threaded and it
> doesn't need to run early.

If useful, I could add the clone wrapper to klibc, but it would have some 
singificant limitations: CLONE_VM would definitely not work (except for with 
CLONE_VFORK) and CLONE_SIGHAND would be cantankerous at best.

However, there are definitely clone flags which would work just fine -- 
CLONE_NEWNS and CLONE_PARENT.

	-hpa