Re: [PATCH] copy_process(): Move fd_install() out of sighand->siglock critical section

[Eric W. Biederman]

Al Viro <viro@xxxxxxxxxxxxxxxxxx> writes:

> On Tue, Feb 08, 2022 at 03:59:06PM -0600, Eric W. Biederman wrote:
>
>> The fd is being installed in the fdtable of the parent process,
>> and the siglock and tasklist_lock are held to protect the child.
>> 
>> 
>> Further fd_install is exposing the fd to userspace where it can be used
>> by the process_madvise and the process_mrelease system calls, from
>> anything that shares the fdtable of the parent thread.  Which means it
>> needs to be guaranteed that kernel_clone will call wake_up_process
>> before it is safe to call fd_install.
>
> You mean "no calling fd_install() until after we are past the last possible
> failure exit, by which point we know that wake_up_process() will eventually
> be called", hopefully?  If so (as I assumed all along), anything downstream
> of
>         if (fatal_signal_pending(current)) {
> 		retval = -EINTR;
> 		goto bad_fork_cancel_cgroup;
> 	}
>
> should be fine...

Except for the problems of calling fd_install under siglock, and
tasklist_lock, which protect nothing and cause lockdep splats.

There may also be assumptions on the task actually being fully setup,
if not today then in a future use pidfd.  So I am not particularly
comfortable with fd_install coming before we drop tasklist_lock.

I was pointing out that to resolve the locking issue we fundamentally
can not move the fd_install earlier, to resolve the locking issues.

Eric