Re: [RFC][v8][PATCH 10/10]: Document clone3() syscall

[Arnd Bergmann <arndbergmann@googlemail.com>]

> 
> > +struct clone_struct {
> > +	u64 flags;
> > +	u64 child_stack;
> 
> u64 seems wrong  on 32 bit platforms. ulong?

That would make it incompatible between 64 bit kernels and
32 bit user space, requiring a wrapper. Better leave it at u64.

> > +	If a pid in the @pids list is 0, the kernel will assign the next
> > +	available pid in the pid namespace, for the process.
> > +
> > +	If a pid in the @pids list is non-zero, the kernel tries to assign
> > +	the specified pid in that namespace.  If that pid is already in use
> > +	by another process, the system call fails with -EBUSY.
> ...
> > +	On failure, clone3() returns -1 and sets 'errno' to one of following
> > +	values (the child process is not created).
> 
> Inconsistent with above. Syscalls really return -ERRCODE, errno is
> glibc magic.

Quite the opposite is true.

The man page describes what the user space sees, which is errno. Returning
-ERRCODE to libc from the kernel is part of the architecture specific
kernel ABI and should not be documented in this place but in the architecture
documentation.

> > +	pid_t pids[] = { 77, 99 };
> > +	struct clone_struct cs;
> > +
> > +	cs.flags = (u64) SIGCHLD;
> > +	cs.child_stack = (u64) setup_child_stack();
> > +	cs.nr_pids = 2;
> > +	cs.parent_tid = 0LL;
> > +	cs.child_tid = 0LL;
> > +
> > +	rc = syscall(__NR_clone3, &cs, pids);
> 
> Hmm, is there reason why pids are not at the end of struct
> clone_struct? Passing most parameters in special structure, then pids
> separately is strange...

I suggested doing that, it's a lot easier to handle fixed length data
structures than an array at the end.

	Arnd <><