Re: [PATCH 32/32] bsd-user: Implement pdfork(2) system call.

[Richard Henderson <richard.henderson@linaro.org>]

On 8/29/23 14:27, Warner Losh wrote:
>      > +    if (put_user_s32(fd, target_fdp)) {
>      > +        return -TARGET_EFAULT;
>      > +    }
> 
>     I *think* this copy belongs in the parent?
> 
> 
> I think that it's copied out in both cases. For normal fork, this would
> be 0 for the pid. However, it appears to return the same FD to both
> the parent and child (see your next comment), so it should be in both
> paths. And even if it returned something different for parent and child
> (which seems unlikely given how the code is setup), we want to return
> the fd each one sees. So either way, I think this code is correct.
> 
>     It's really hard to follow the path of new
>     process creation within the freebsd kernel.
> 
> 
> Agreed.

I think that the child never returns from do_fork.  The child pid == 0 happens as part of 
do_fork or vm_forkproc or somesuch, but the new process definitely begins life at fork_return.

Therefore only the parent passes returns from fork1 to set *fdp.


r~