[PATCH v3 net-next] udp: introduce __sk_mem_schedule() usage

[PATCH v3 net-next] udp: introduce __sk_mem_schedule() usage

[Jason Xing]

From: Jason Xing <kernelxing@tencent.com>

Keep the accounting schema consistent across different protocols
with __sk_mem_schedule(). Besides, it adjusts a little bit on how
to calculate forward allocated memory compared to before. After
applied this patch, we could avoid receive path scheduling extra
amount of memory.

Link: https://lore.kernel.org/lkml/20230221110344.82818-1-kerneljasonxing@gmail.com/
Signed-off-by: Jason Xing <kernelxing@tencent.com>
---
v3:
1) get rid of inline suggested by Simon Horman

v2:
1) change the title and body message
2) use __sk_mem_schedule() instead suggested by Paolo Abeni
---
 net/ipv4/udp.c | 31 ++++++++++++++++++-------------
 1 file changed, 18 insertions(+), 13 deletions(-)

diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index c605d171eb2d..60473781933c 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1531,10 +1531,23 @@ static void busylock_release(spinlock_t *busy)
 		spin_unlock(busy);
 }
 
+static int udp_rmem_schedule(struct sock *sk, int size)
+{
+	int delta;
+
+	delta = size - sk->sk_forward_alloc;
+	if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
+		return -ENOBUFS;
+
+	sk->sk_forward_alloc -= size;
+
+	return 0;
+}
+
 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
 {
 	struct sk_buff_head *list = &sk->sk_receive_queue;
-	int rmem, delta, amt, err = -ENOMEM;
+	int rmem, err = -ENOMEM;
 	spinlock_t *busy = NULL;
 	int size;
 
@@ -1567,20 +1580,12 @@ int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
 		goto uncharge_drop;
 
 	spin_lock(&list->lock);
-	if (size >= sk->sk_forward_alloc) {
-		amt = sk_mem_pages(size);
-		delta = amt << PAGE_SHIFT;
-		if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
-			err = -ENOBUFS;
-			spin_unlock(&list->lock);
-			goto uncharge_drop;
-		}
-
-		sk->sk_forward_alloc += delta;
+	err = udp_rmem_schedule(sk, size);
+	if (err) {
+		spin_unlock(&list->lock);
+		goto uncharge_drop;
 	}
 
-	sk->sk_forward_alloc -= size;
-
 	/* no need to setup a destructor, we will explicitly release the
 	 * forward allocated memory on dequeue
 	 */
-- 
2.37.3

Re: [PATCH v3 net-next] udp: introduce __sk_mem_schedule() usage

[Paolo Abeni]

On Tue, 2023-03-07 at 09:56 +0800, Jason Xing wrote:
> From: Jason Xing <kernelxing@tencent.com>
> 
> Keep the accounting schema consistent across different protocols
> with __sk_mem_schedule(). Besides, it adjusts a little bit on how
> to calculate forward allocated memory compared to before. After
> applied this patch, we could avoid receive path scheduling extra
> amount of memory.
> 
> Link: https://lore.kernel.org/lkml/20230221110344.82818-1-kerneljasonxing@gmail.com/
> Signed-off-by: Jason Xing <kernelxing@tencent.com>
> ---
> v3:
> 1) get rid of inline suggested by Simon Horman
> 
> v2:
> 1) change the title and body message
> 2) use __sk_mem_schedule() instead suggested by Paolo Abeni
> ---
>  net/ipv4/udp.c | 31 ++++++++++++++++++-------------
>  1 file changed, 18 insertions(+), 13 deletions(-)
> 
> diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
> index c605d171eb2d..60473781933c 100644
> --- a/net/ipv4/udp.c
> +++ b/net/ipv4/udp.c
> @@ -1531,10 +1531,23 @@ static void busylock_release(spinlock_t *busy)
>  		spin_unlock(busy);
>  }
>  
> +static int udp_rmem_schedule(struct sock *sk, int size)
> +{
> +	int delta;
> +
> +	delta = size - sk->sk_forward_alloc;
> +	if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
> +		return -ENOBUFS;
> +
> +	sk->sk_forward_alloc -= size;

I think it's better if you maintain the above statement outside of this
helper: it's a bit confusing that rmem_schedule() actually consumes fwd
memory.

Side note

Cheers,

Paolo

Re: [PATCH v3 net-next] udp: introduce __sk_mem_schedule() usage

[Jason Xing]

On Tue, Mar 7, 2023 at 10:55 PM Paolo Abeni <pabeni@redhat.com> wrote:
>
> On Tue, 2023-03-07 at 09:56 +0800, Jason Xing wrote:
> > From: Jason Xing <kernelxing@tencent.com>
> >
> > Keep the accounting schema consistent across different protocols
> > with __sk_mem_schedule(). Besides, it adjusts a little bit on how
> > to calculate forward allocated memory compared to before. After
> > applied this patch, we could avoid receive path scheduling extra
> > amount of memory.
> >
> > Link: https://lore.kernel.org/lkml/20230221110344.82818-1-kerneljasonxing@gmail.com/
> > Signed-off-by: Jason Xing <kernelxing@tencent.com>
> > ---
> > v3:
> > 1) get rid of inline suggested by Simon Horman
> >
> > v2:
> > 1) change the title and body message
> > 2) use __sk_mem_schedule() instead suggested by Paolo Abeni
> > ---
> >  net/ipv4/udp.c | 31 ++++++++++++++++++-------------
> >  1 file changed, 18 insertions(+), 13 deletions(-)
> >
> > diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
> > index c605d171eb2d..60473781933c 100644
> > --- a/net/ipv4/udp.c
> > +++ b/net/ipv4/udp.c
> > @@ -1531,10 +1531,23 @@ static void busylock_release(spinlock_t *busy)
> >               spin_unlock(busy);
> >  }
> >
> > +static int udp_rmem_schedule(struct sock *sk, int size)
> > +{
> > +     int delta;
> > +
> > +     delta = size - sk->sk_forward_alloc;
> > +     if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
> > +             return -ENOBUFS;
> > +
> > +     sk->sk_forward_alloc -= size;
>
> I think it's better if you maintain the above statement outside of this
> helper: it's a bit confusing that rmem_schedule() actually consumes fwd
> memory.

It does make sense.

Thanks,
Jason

>
> Side note
>
> Cheers,
>
> Paolo
>