Re: [PATCH] release_task: kill the no longer needed get/put_pid(thread_pid)

[Mateusz Guzik]

On Mon, Apr 14, 2025 at 9:45 PM Christian Brauner <brauner@xxxxxxxxxx> wrote:
>
> On Mon, Apr 14, 2025 at 09:39:47PM +0200, Christian Brauner wrote:
> > On Fri, Apr 11, 2025 at 02:18:57PM +0200, Oleg Nesterov wrote:
> > > After the commit 7903f907a2260 ("pid: perform free_pid() calls outside
> > > of tasklist_lock") __unhash_process() -> detach_pid() no longer calls
> > > free_pid(), proc_flush_pid() can just use p->thread_pid without the
> > > now pointless get_pid() + put_pid().
> > >
> > > Signed-off-by: Oleg Nesterov <oleg@xxxxxxxxxx>
> > > ---
> > >  kernel/exit.c | 7 ++-----
> > >  1 file changed, 2 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/kernel/exit.c b/kernel/exit.c
> > > index 1b51dc099f1e..96d639383f86 100644
> > > --- a/kernel/exit.c
> > > +++ b/kernel/exit.c
> > > @@ -239,7 +239,6 @@ void release_task(struct task_struct *p)
> > >  {
> > >     struct release_task_post post;
> > >     struct task_struct *leader;
> > > -   struct pid *thread_pid;
> > >     int zap_leader;
> > >  repeat:
> > >     memset(&post, 0, sizeof(post));
> > > @@ -253,8 +252,6 @@ void release_task(struct task_struct *p)
> > >     pidfs_exit(p);
> > >     cgroup_release(p);
> > >
> > > -   thread_pid = get_pid(p->thread_pid);
> > > -
> > >     write_lock_irq(&tasklist_lock);
> > >     ptrace_release_task(p);
> > >     __exit_signal(&post, p);
> > > @@ -282,8 +279,8 @@ void release_task(struct task_struct *p)
> > >     }
> > >
> > >     write_unlock_irq(&tasklist_lock);
> > > -   proc_flush_pid(thread_pid);
> > > -   put_pid(thread_pid);
> > > +   /* p->thread_pid can't go away until free_pids() below */
> > > +   proc_flush_pid(p->thread_pid);
> >
> > This cannot work though, right?
> > Because after __unhash_process() p->thread_pid may be NULL:
> >
> > __unhash_process()
> > -> detach_pid()
> >    -> __change_pid()
> >       {
> >       struct pid **pid_ptr, *pid;
> >
> >       pid_ptr = task_pid_ptr(task, type);
> >
> >       *pid_ptr = NULL;
> >
> >       for (tmp = PIDTYPE_MAX; --tmp >= 0; )
> >               if (pid_has_task(pid, tmp)) /* will be false if @group_dead is true
> >                       return;
> >
> >       WARN_ON(pids[type]);
> >       pids[type] = pid;
> >       }
> >
> > so this needs:
> >
> > diff --git a/kernel/exit.c b/kernel/exit.c
> > index e6132ebdaed4..9232c4c684e9 100644
> > --- a/kernel/exit.c
> > +++ b/kernel/exit.c
> > @@ -244,6 +244,7 @@ void release_task(struct task_struct *p)
> >  {
> >         struct release_task_post post;
> >         struct task_struct *leader;
> > +       struct pid *thread_pid = task_pid(p);
> >         int zap_leader;
> >  repeat:
> >         memset(&post, 0, sizeof(post));
> > @@ -285,7 +286,7 @@ void release_task(struct task_struct *p)
> >
> >         write_unlock_irq(&tasklist_lock);
> >         /* p->thread_pid can't go away until free_pids() below */
> > -       proc_flush_pid(p->thread_pid);
> > +       proc_flush_pid(thread_pid);
> >         add_device_randomness(&p->se.sum_exec_runtime,
> >                               sizeof(p->se.sum_exec_runtime));
> >         free_pids(post.pids);
> >
> > I've folded this diff into your patch, Oleg. Let me know if you see any
> > additional issues with this.
>
> The task_pid() needs to be moved after the repeat label. I'm appending
> the full patch I applied.

oh heh, ack on that

but while here perhaps a small stylistic cleanup: move
add_device_randomness before or after proc_flush_pid + free_pids,
instead of if being in-between

-- 
Mateusz Guzik <mjguzik gmail.com>