[PATCH] ksmbd: server: avoid busy polling in accept loop

[PATCH] ksmbd: server: avoid busy polling in accept loop

[Qingfang Deng]

The ksmbd listener thread was using busy waiting on a listening socket by
calling kernel_accept() with SOCK_NONBLOCK and retrying every 100ms on
-EAGAIN. Since this thread is dedicated to accepting new connections,
there is no need for non-blocking mode.

Switch to a blocking accept() call instead, allowing the thread to sleep
until a new connection arrives. This avoids unnecessary wakeups and CPU
usage.

Also remove:
  - TCP_NODELAY, which has no effect on a listening socket.
  - sk_rcvtimeo and sk_sndtimeo assignments, which only caused accept()
    to return -EAGAIN prematurely.

Fixes: 0626e6641f6b ("cifsd: add server handler for central processing and tranport layers")
Signed-off-by: Qingfang Deng <dqfext@gmail.com>
---
 fs/smb/server/transport_tcp.c | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)

diff --git a/fs/smb/server/transport_tcp.c b/fs/smb/server/transport_tcp.c
index 7a1e3dcc2cde..57a6aa98e7de 100644
--- a/fs/smb/server/transport_tcp.c
+++ b/fs/smb/server/transport_tcp.c
@@ -46,11 +46,6 @@ static struct interface *alloc_iface(char *ifname);
 #define TCP_TRANS(t)	((struct tcp_transport *)container_of(t, \
 				struct tcp_transport, transport))
 
-static inline void ksmbd_tcp_nodelay(struct socket *sock)
-{
-	tcp_sock_set_nodelay(sock->sk);
-}
-
 static inline void ksmbd_tcp_reuseaddr(struct socket *sock)
 {
 	sock_set_reuseaddr(sock->sk);
@@ -241,15 +236,10 @@ static int ksmbd_kthread_fn(void *p)
 			mutex_unlock(&iface->sock_release_lock);
 			break;
 		}
-		ret = kernel_accept(iface->ksmbd_socket, &client_sk,
-				    SOCK_NONBLOCK);
+		ret = kernel_accept(iface->ksmbd_socket, &client_sk, 0);
 		mutex_unlock(&iface->sock_release_lock);
-		if (ret) {
-			if (ret == -EAGAIN)
-				/* check for new connections every 100 msecs */
-				schedule_timeout_interruptible(HZ / 10);
+		if (ret)
 			continue;
-		}
 
 		if (!server_conf.max_ip_connections)
 			goto skip_max_ip_conns_limit;
@@ -455,10 +445,6 @@ static void tcp_destroy_socket(struct socket *ksmbd_socket)
 	if (!ksmbd_socket)
 		return;
 
-	/* set zero to timeout */
-	ksmbd_tcp_rcv_timeout(ksmbd_socket, 0);
-	ksmbd_tcp_snd_timeout(ksmbd_socket, 0);
-
 	ret = kernel_sock_shutdown(ksmbd_socket, SHUT_RDWR);
 	if (ret)
 		pr_err("Failed to shutdown socket: %d\n", ret);
@@ -505,7 +491,6 @@ static int create_socket(struct interface *iface)
 		release_sock(ksmbd_socket->sk);
 	}
 
-	ksmbd_tcp_nodelay(ksmbd_socket);
 	ksmbd_tcp_reuseaddr(ksmbd_socket);
 
 	ret = sock_setsockopt(ksmbd_socket,
@@ -529,9 +514,6 @@ static int create_socket(struct interface *iface)
 		goto out_error;
 	}
 
-	ksmbd_socket->sk->sk_rcvtimeo = KSMBD_TCP_RECV_TIMEOUT;
-	ksmbd_socket->sk->sk_sndtimeo = KSMBD_TCP_SEND_TIMEOUT;
-
 	ret = kernel_listen(ksmbd_socket, KSMBD_SOCKET_BACKLOG);
 	if (ret) {
 		pr_err("Port listen() error: %d\n", ret);
-- 
2.43.0

Re: [PATCH] ksmbd: server: avoid busy polling in accept loop

[Namjae Jeon]

On Thu, Oct 30, 2025 at 3:47 PM Qingfang Deng <dqfext@gmail.com> wrote:
>
> The ksmbd listener thread was using busy waiting on a listening socket by
> calling kernel_accept() with SOCK_NONBLOCK and retrying every 100ms on
> -EAGAIN. Since this thread is dedicated to accepting new connections,
> there is no need for non-blocking mode.
>
> Switch to a blocking accept() call instead, allowing the thread to sleep
> until a new connection arrives. This avoids unnecessary wakeups and CPU
> usage.
>
> Also remove:
>   - TCP_NODELAY, which has no effect on a listening socket.
>   - sk_rcvtimeo and sk_sndtimeo assignments, which only caused accept()
>     to return -EAGAIN prematurely.
>
> Fixes: 0626e6641f6b ("cifsd: add server handler for central processing and tranport layers")
> Signed-off-by: Qingfang Deng <dqfext@gmail.com>
Applied it to #ksmbd-for-next-next.
Thanks!

Re: [PATCH] ksmbd: server: avoid busy polling in accept loop

[Qingfang Deng]

Hi Namjae,

On Thu, Oct 30, 2025 at 4:11 PM Namjae Jeon <linkinjeon@kernel.org> wrote:
> > Fixes: 0626e6641f6b ("cifsd: add server handler for central processing and tranport layers")
> > Signed-off-by: Qingfang Deng <dqfext@gmail.com>
> Applied it to #ksmbd-for-next-next.
> Thanks!

I just found that this depends on another commit which is not in
kernel versions earlier than v6.1:
a7c01fa93aeb ("signal: break out of wait loops on kthread_stop()")

With the current Fixes tag, this commit will be backported to v5.15
automatically. But without said commit, kthread_stop() cannot wake up
a blocking kernel_accept().
Should I change the Fixes tag, or inform linux-stable not to backport
this patch to v5.15?

+Cc: Jason, Sasha, and GregKH

Regards,
Qingfang

Re: [PATCH] ksmbd: server: avoid busy polling in accept loop

[Greg Kroah-Hartman]

On Fri, Oct 31, 2025 at 03:32:06PM +0800, Qingfang Deng wrote:
> Hi Namjae,
> 
> On Thu, Oct 30, 2025 at 4:11 PM Namjae Jeon <linkinjeon@kernel.org> wrote:
> > > Fixes: 0626e6641f6b ("cifsd: add server handler for central processing and tranport layers")
> > > Signed-off-by: Qingfang Deng <dqfext@gmail.com>
> > Applied it to #ksmbd-for-next-next.
> > Thanks!
> 
> I just found that this depends on another commit which is not in
> kernel versions earlier than v6.1:
> a7c01fa93aeb ("signal: break out of wait loops on kthread_stop()")
> 
> With the current Fixes tag, this commit will be backported to v5.15
> automatically. But without said commit, kthread_stop() cannot wake up
> a blocking kernel_accept().
> Should I change the Fixes tag, or inform linux-stable not to backport
> this patch to v5.15?

Email stable@vger.kernel.org when it lands in Linus's tree to not
backport it that far.

thanks,

greg k-h

Re: [PATCH] ksmbd: server: avoid busy polling in accept loop

[Qingfang Deng]

On Fri, Oct 31, 2025 at 3:44 PM Greg Kroah-Hartman
<gregkh@linuxfoundation.org> wrote:
>
> On Fri, Oct 31, 2025 at 03:32:06PM +0800, Qingfang Deng wrote:
> > Hi Namjae,
> >
> > On Thu, Oct 30, 2025 at 4:11 PM Namjae Jeon <linkinjeon@kernel.org> wrote:
> > > > Fixes: 0626e6641f6b ("cifsd: add server handler for central processing and tranport layers")
> > > > Signed-off-by: Qingfang Deng <dqfext@gmail.com>
> > > Applied it to #ksmbd-for-next-next.
> > > Thanks!
> >
> > I just found that this depends on another commit which is not in
> > kernel versions earlier than v6.1:
> > a7c01fa93aeb ("signal: break out of wait loops on kthread_stop()")
> >
> > With the current Fixes tag, this commit will be backported to v5.15
> > automatically. But without said commit, kthread_stop() cannot wake up
> > a blocking kernel_accept().
> > Should I change the Fixes tag, or inform linux-stable not to backport
> > this patch to v5.15?
>
> Email stable@vger.kernel.org when it lands in Linus's tree to not
> backport it that far.

Noted. Thanks!

Re: [PATCH] ksmbd: server: avoid busy polling in accept loop

[Namjae Jeon]

On Fri, Oct 31, 2025 at 4:49 PM Qingfang Deng <dqfext@gmail.com> wrote:
>
> On Fri, Oct 31, 2025 at 3:44 PM Greg Kroah-Hartman
> <gregkh@linuxfoundation.org> wrote:
> >
> > On Fri, Oct 31, 2025 at 03:32:06PM +0800, Qingfang Deng wrote:
> > > Hi Namjae,
> > >
> > > On Thu, Oct 30, 2025 at 4:11 PM Namjae Jeon <linkinjeon@kernel.org> wrote:
> > > > > Fixes: 0626e6641f6b ("cifsd: add server handler for central processing and tranport layers")
> > > > > Signed-off-by: Qingfang Deng <dqfext@gmail.com>
> > > > Applied it to #ksmbd-for-next-next.
> > > > Thanks!
> > >
> > > I just found that this depends on another commit which is not in
> > > kernel versions earlier than v6.1:
> > > a7c01fa93aeb ("signal: break out of wait loops on kthread_stop()")
> > >
> > > With the current Fixes tag, this commit will be backported to v5.15
> > > automatically. But without said commit, kthread_stop() cannot wake up
> > > a blocking kernel_accept().
> > > Should I change the Fixes tag, or inform linux-stable not to backport
> > > this patch to v5.15?
> >
> > Email stable@vger.kernel.org when it lands in Linus's tree to not
> > backport it that far.
>
> Noted. Thanks!
I think it would be better to use the stable tag + kernel version tag
instead of the fixes tag.

Cc: stable@vger.kernel.org # v6.1+

Also, I don't think this patch necessarily needs to be merged into the
stable kernels.
Thanks.