On (22/12/27 23:59), Namjae Jeon wrote:
[..]
> csin = KSMBD_TCP_PEER_SOCKADDR(KSMBD_TRANS(t)->conn);
> if (kernel_getpeername(client_sk, csin) < 0) {
> @@ -239,6 +243,17 @@ static int ksmbd_kthread_fn(void *p)
> continue;
> }
>
> + if (server_conf.max_connections) {
> + if (atomic_read(&active_num_conn) >= server_conf.max_connections) {
> + pr_info("Limit the maximum number of connections(%u)\n",
> + atomic_read(&active_num_conn));
> + sock_release(client_sk);
> + continue;
> + }
> +
> + atomic_inc(&active_num_conn);
> + }
This has to be one atomic op:
if (atomic_inc_return() >= max_connections) {
atomic_dec
sock_release
continue
}
Otherwise it's racy and it's possible to have more active
connections than the limit.
I'd also note that pr_info() is risky there, it would be safer
to use rate-limited printk().
