Re: [PATCH v2 net-next 5/5] tcp: Introduce optional per-netns ehash.

[Kuniyuki Iwashima <kuniyu@amazon.com>]

From:   Eric Dumazet <edumazet@google.com>
Date:   Mon, 29 Aug 2022 15:59:04 -0700
> On Mon, Aug 29, 2022 at 9:21 AM Kuniyuki Iwashima <kuniyu@amazon.com> wrote:
> >
> > The more sockets we have in the hash table, the longer we spend looking
> > up the socket.  While running a number of small workloads on the same
> > host, they penalise each other and cause performance degradation.>
> 
> ...
> > +static int proc_tcp_child_ehash_entries(struct ctl_table *table, int write,
> > +                                       void *buffer, size_t *lenp, loff_t *ppos)
> > +{
> > +       unsigned int tcp_child_ehash_entries;
> > +       int ret;
> > +
> > +       ret = proc_douintvec_minmax(table, write, buffer, lenp, ppos);
> > +       if (!write || ret)
> > +               return ret;
> > +
> > +       tcp_child_ehash_entries = READ_ONCE(*(unsigned int *)table->data);
> > +       if (tcp_child_ehash_entries)
> > +               tcp_child_ehash_entries = roundup_pow_of_two(tcp_child_ehash_entries);
> 
> This is not thread safe.

Oh, I didn't know that.
Thank you for pointing out!

> 
> You could simply perform the roundup_pow_of_two() elsewhere,
> eg in tcp_set_hashinfo() (and leave the sysctl as set by the user)

Will do so and update the doc and changelog.


> 
> > +
> > +       WRITE_ONCE(*(unsigned int *)table->data, tcp_child_ehash_entries);
> > +
> > +       return 0;
> > +}
> > +