Re: [PATCH 5/9] signal/group_exit: Use start_group_exit in place of do_group_exit

[Kees Cook <keescook@chromium.org>]

On Thu, Jun 24, 2021 at 02:01:40PM -0500, Eric W. Biederman wrote:
> 
> Make thread exiting uniform by causing all threads to pass through
> get_signal when they are exiting.  This simplifies the analysis
> of sychronization during exit and guarantees that all full set
> of registers will be available for ptrace to examine for
> threads that stop at PTRACE_EVENT_EXIT.

Yeah, cool. I do like making the process lifetime more sensible here. It
always threw me that do_exit*() just stopped execution. :)

For future me, can you add a comment on start_group_exit() that mentions
where final process death happens?

> 
> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
> ---
>  kernel/exit.c    | 4 ++--
>  kernel/seccomp.c | 2 +-
>  2 files changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/exit.c b/kernel/exit.c
> index fd1c04193e18..921519d80b56 100644
> --- a/kernel/exit.c
> +++ b/kernel/exit.c
> @@ -931,8 +931,8 @@ do_group_exit(int exit_code)
>   */
>  SYSCALL_DEFINE1(exit_group, int, error_code)
>  {
> -	do_group_exit((error_code & 0xff) << 8);
> -	/* NOTREACHED */
> +	start_group_exit((error_code & 0xff) << 8);
> +	/* get_signal will call do_exit */
>  	return 0;

"0" feels weird here, but I can't think of any better "fail closed"
error code here.

>  }
>  
> diff --git a/kernel/seccomp.c b/kernel/seccomp.c
> index 5301eca670a0..b1c06fd1b205 100644
> --- a/kernel/seccomp.c
> +++ b/kernel/seccomp.c
> @@ -1250,7 +1250,7 @@ static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
>  			if (action == SECCOMP_RET_KILL_THREAD)
>  				do_exit(SIGSYS);
>  			else
> -				do_group_exit(SIGSYS);
> +				start_group_exit(SIGSYS);

This could use a similar comment to the syscall's comment, just so I
don't panic when I read this code in like 3 years. ;)

Otherwise, yeah, looks good.

-Kees

>  		}
>  		return -1;
>  	}
> -- 
> 2.20.1
> 

-- 
Kees Cook