Re: [PATCH v3 net-next 04/14] af_unix: Bulk update unix_tot_inflight/unix_inflight when queuing skb.

[Kuniyuki Iwashima <kuniyu@amazon.com>]

From: Paolo Abeni <pabeni@redhat.com>
Date: Tue, 27 Feb 2024 11:47:23 +0100
> On Fri, 2024-02-23 at 13:39 -0800, Kuniyuki Iwashima wrote:
> > diff --git a/net/unix/garbage.c b/net/unix/garbage.c
> > index 96d0b1db3638..e8fe08796d02 100644
> > --- a/net/unix/garbage.c
> > +++ b/net/unix/garbage.c
> > @@ -148,6 +148,7 @@ static void unix_free_vertices(struct scm_fp_list *fpl)
> >  }
> >  
> >  DEFINE_SPINLOCK(unix_gc_lock);
> > +unsigned int unix_tot_inflight;
> >  
> >  void unix_add_edges(struct scm_fp_list *fpl, struct unix_sock *receiver)
> >  {
> > @@ -172,7 +173,10 @@ void unix_add_edges(struct scm_fp_list *fpl, struct unix_sock *receiver)
> >  		unix_add_edge(fpl, edge);
> >  	} while (i < fpl->count_unix);
> >  
> > +	WRITE_ONCE(unix_tot_inflight, unix_tot_inflight + fpl->count_unix);
> >  out:
> > +	WRITE_ONCE(fpl->user->unix_inflight, fpl->user->unix_inflight + fpl->count);
> 
> I'm unsure if later patches will shed some light, but why the above
> statement is placed _after_ the 'out' label? fpl->count will be 0 in
> such path, and the updated not needed. Why don't you place it before
> the mentioned label?

fpl->count is the total number of fds in skb, and fpl->count_unix
is the number of AF_UNIX fds.

So, we could reach the out: label if we pass no AF_UNIX fd but
other fds, then we count the number for each user to use in
too_many_unix_fds().

Before this change, unix_inflight() and unix_notinflight() did the
same but incremented/decremented one by one.


> 
> > +
> >  	spin_unlock(&unix_gc_lock);
> >  
> >  	fpl->inflight = true;
> > @@ -195,7 +199,10 @@ void unix_del_edges(struct scm_fp_list *fpl)
> >  		unix_del_edge(fpl, edge);
> >  	} while (i < fpl->count_unix);
> >  
> > +	WRITE_ONCE(unix_tot_inflight, unix_tot_inflight - fpl->count_unix);
> >  out:
> > +	WRITE_ONCE(fpl->user->unix_inflight, fpl->user->unix_inflight - fpl->count);
> 
> Same question here.
> 
> Thanks!
> 
> Paolo