Re: [RFC] lsm: fs: Use i_callback to free i_security in RCU callback

[Song Liu <song@xxxxxxxxxx>]

On Tue, Dec 17, 2024 at 9:38 AM Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> >  - Let pipe free inode from a RCU callback.
>
> ... which hurts the systems with LSM crap disabled.
> NAK.

How do we measure the overhead in such cases? AFAICT,
the overhead is very small:

1. Many (most) systems have some LSM enabled anyway.
2. pipe create/release is not on any hot path. On a busy system
  with 176 CPUs, I measured ~30 pipe create/release per second.
3. The overhead of a rcu callback is small.

Given these measures, I don't think "hurts the system without LSM"
justifies 2 extra pointers per inode.

Thanks,
Song