Re: [GIT PULL] KVM changes for Linux 6.14

[Oleg Nesterov <oleg@redhat.com>]

On 01/25, Linus Torvalds wrote:
>
> Keith pinpointed the user space logic to fork_remap():
>
>    https://github.com/google/minijail/blob/main/rust/minijail/src/lib.rs#L987
>
> and honestly, I do think it makes sense for user space to ask "am I
> single-threaded" (which is presumably the thing that breaks), and the
> code for that is pretty simple:
>
>   fn is_single_threaded() -> io::Result<bool> {
>       match count_dir_entries("/proc/self/task") {
>           Ok(1) => Ok(true),
>           Ok(_) => Ok(false),
>           Err(e) => Err(e),
>       }
>   }
>
> and I really don't think user space is "wrong".
>
> So the fact that a kernel helper thread that runs async in the
> background and does random background infrastructure things that do
> not really affect user space should probably simply not break this
> kind of simple (and admittedly simplistic) user space logic.
>
> Should we just add some flag to say "don't show this thread in this
> context"?

Not sure I understand... Looking at is_single_threaded() above I guess
something like below should work (incomplete, in particular we need to
chang first_tid() as well).

But a PF_HIDDEN sub-thread will still be visible via /proc/$pid_of_PF_HIDDEN

> We obviously still want to see it for management purposes,
> so it's not like the thing should be entirely invisible,

Can you explain?

Oleg.


--- x/include/linux/sched.h
+++ x/include/linux/sched.h
@@ -1685,7 +1685,7 @@ extern struct pid *cad_pid;
 #define PF_USED_MATH		0x00002000	/* If unset the fpu must be initialized before use */
 #define PF_USER_WORKER		0x00004000	/* Kernel thread cloned from userspace thread */
 #define PF_NOFREEZE		0x00008000	/* This thread should not be frozen */
-#define PF__HOLE__00010000	0x00010000
+#define PF_HIDDEN		0x00010000
 #define PF_KSWAPD		0x00020000	/* I am kswapd */
 #define PF_MEMALLOC_NOFS	0x00040000	/* All allocations inherit GFP_NOFS. See memalloc_nfs_save() */
 #define PF_MEMALLOC_NOIO	0x00080000	/* All allocations inherit GFP_NOIO. See memalloc_noio_save() */
--- x/include/linux/sched/task.h
+++ x/include/linux/sched/task.h
@@ -31,6 +31,7 @@ struct kernel_clone_args {
 	u32 io_thread:1;
 	u32 user_worker:1;
 	u32 no_files:1;
+	u32 hidden:1;
 	unsigned long stack;
 	unsigned long stack_size;
 	unsigned long tls;
--- x/kernel/fork.c
+++ x/kernel/fork.c
@@ -2237,6 +2237,8 @@ __latent_entropy struct task_struct *cop
 	}
 	if (args->io_thread)
 		p->flags |= PF_IO_WORKER;
+	if (args->hidden)
+		p->flags |= PF_HIDDEN;
 
 	if (args->name)
 		strscpy_pad(p->comm, args->name, sizeof(p->comm));
--- x/kernel/vhost_task.c
+++ x/kernel/vhost_task.c
@@ -117,7 +117,7 @@ EXPORT_SYMBOL_GPL(vhost_task_stop);
  */
 struct vhost_task *vhost_task_create(bool (*fn)(void *),
 				     void (*handle_sigkill)(void *), void *arg,
-				     const char *name)
+				     bool hidden, const char *name)
 {
 	struct kernel_clone_args args = {
 		.flags		= CLONE_FS | CLONE_UNTRACED | CLONE_VM |
@@ -125,6 +125,7 @@ struct vhost_task *vhost_task_create(boo
 		.exit_signal	= 0,
 		.fn		= vhost_task_fn,
 		.name		= name,
+		.hidden		= hidden,
 		.user_worker	= 1,
 		.no_files	= 1,
 	};
--- x/fs/proc/base.c
+++ x/fs/proc/base.c
@@ -3906,9 +3906,12 @@ static struct task_struct *next_tid(stru
 	struct task_struct *pos = NULL;
 	rcu_read_lock();
 	if (pid_alive(start)) {
-		pos = __next_thread(start);
-		if (pos)
-			get_task_struct(pos);
+		for (pos = start; (pos = __next_thread(pos)); ) {
+			if (!(pos->flags & PF_HIDDEN)) {
+				get_task_struct(pos);
+				break;
+			}
+		}
 	}
 	rcu_read_unlock();
 	put_task_struct(start);