Re: [GIT PULL] pidfd updates

[Linus Torvalds]

On Thu, Apr 27, 2023 at 12:39 AM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> int delayed_dup(struct file *file, unsigned flags)

Ok, this is strange. Let me think about it.

But even without thinking about it, this part I hate:

>         struct delayed_dup *p = kmalloc(sizeof(struct delayed_dup), GFP_KERNEL);

Sure, if this is only used in unimportant code where performance
doesn't matter, doing a kmalloc is fine.

But if that is the only use, I think this is too subtle an interface.

Could we instead limit it to "we only have one pending delayed dup",
and make this all be more like the restart-block thing, and be part of
struct task_struct?

I think it's conceptually quite similar to restart_block, ie a "some
pending system call state" thing.

(Obviously it's entirely different in details).

          Linus