* [PATCH net-next] net: sock: adapt SOCK_MIN_RCVBUF and SOCK_MIN_SNDBUF
@ 2013-06-19 9:18 Daniel Borkmann
2013-06-19 9:51 ` Eric Dumazet
0 siblings, 1 reply; 5+ messages in thread
From: Daniel Borkmann @ 2013-06-19 9:18 UTC (permalink / raw)
To: davem; +Cc: netdev
The current situation is that SOCK_MIN_RCVBUF is 2048 + sizeof(struct sk_buff))
while SOCK_MIN_SNDBUF is 2048. Since in both cases, skb->truesize is used for
sk_{r,w}mem_alloc accounting, we should have both sizes equal and adjusted
through the macro SKB_TRUESIZE(), which is also used elsewhere to adjust sk
buffer sizes. The minor adaption in sk_stream_moderate_sndbuf() is to silence
a warning by using a typed max macro, as similarly done in SOCK_MIN_RCVBUF
occurences, that would appear otherwise.
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
---
include/net/sock.h | 11 ++++++-----
1 file changed, 6 insertions(+), 5 deletions(-)
diff --git a/include/net/sock.h b/include/net/sock.h
index ac8e181..189ef98 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2045,18 +2045,19 @@ static inline void sk_wake_async(struct sock *sk, int how, int band)
sock_wake_async(sk->sk_socket, how, band);
}
-#define SOCK_MIN_SNDBUF 2048
/*
- * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
- * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
+ * Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
+ * need sizeof(sk_buff) + sizeof(skb_shared_info) + MTU + padding, unless
+ * net driver perform copybreak.
*/
-#define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
+#define SOCK_MIN_RCVBUF SKB_TRUESIZE(2048)
+#define SOCK_MIN_SNDBUF SKB_TRUESIZE(2048)
static inline void sk_stream_moderate_sndbuf(struct sock *sk)
{
if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
- sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
+ sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
}
}
--
1.7.11.7
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH net-next] net: sock: adapt SOCK_MIN_RCVBUF and SOCK_MIN_SNDBUF
2013-06-19 9:18 [PATCH net-next] net: sock: adapt SOCK_MIN_RCVBUF and SOCK_MIN_SNDBUF Daniel Borkmann
@ 2013-06-19 9:51 ` Eric Dumazet
2013-06-19 9:57 ` Daniel Borkmann
2013-06-19 9:58 ` Eric Dumazet
0 siblings, 2 replies; 5+ messages in thread
From: Eric Dumazet @ 2013-06-19 9:51 UTC (permalink / raw)
To: Daniel Borkmann; +Cc: davem, netdev
On Wed, 2013-06-19 at 11:18 +0200, Daniel Borkmann wrote:
> The current situation is that SOCK_MIN_RCVBUF is 2048 + sizeof(struct sk_buff))
> while SOCK_MIN_SNDBUF is 2048. Since in both cases, skb->truesize is used for
> sk_{r,w}mem_alloc accounting, we should have both sizes equal and adjusted
> through the macro SKB_TRUESIZE(), which is also used elsewhere to adjust sk
> buffer sizes. The minor adaption in sk_stream_moderate_sndbuf() is to silence
> a warning by using a typed max macro, as similarly done in SOCK_MIN_RCVBUF
> occurences, that would appear otherwise.
>
> Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
> ---
> include/net/sock.h | 11 ++++++-----
> 1 file changed, 6 insertions(+), 5 deletions(-)
>
> diff --git a/include/net/sock.h b/include/net/sock.h
> index ac8e181..189ef98 100644
> --- a/include/net/sock.h
> +++ b/include/net/sock.h
> @@ -2045,18 +2045,19 @@ static inline void sk_wake_async(struct sock *sk, int how, int band)
> sock_wake_async(sk->sk_socket, how, band);
> }
>
> -#define SOCK_MIN_SNDBUF 2048
> /*
> - * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
> - * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
> + * Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
> + * need sizeof(sk_buff) + sizeof(skb_shared_info) + MTU + padding, unless
> + * net driver perform copybreak.
> */
> -#define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
> +#define SOCK_MIN_RCVBUF SKB_TRUESIZE(2048)
> +#define SOCK_MIN_SNDBUF SKB_TRUESIZE(2048)
>
>
> static inline void sk_stream_moderate_sndbuf(struct sock *sk)
> {
> if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
> sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
> - sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
> + sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
> }
> }
>
Funny you send this patch, because I prepared a similar patch
yesterday ;)
My motivation was a bit different, because we hit a (small) regression
here in Google for some applications setting low SO_SNDBUF/SO_RCVBUF
values, because of new TCP needs :
Minimal skb truesize in transmit path is indeed SKB_TRUESIZE(2048) after
commit f07d960df33c5aef ("tcp: avoid frag allocation for small frames")
And tcp sendmsg() tries to limit skb size to half the congestion window,
meaning we try to build two skbs at minimum.
So I believe that we need :
/* TCP works better if we can build two skbs at minimum */
#define SOCK_MIN_SNDBUF (2 * SKB_TRUESIZE(2048))
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH net-next] net: sock: adapt SOCK_MIN_RCVBUF and SOCK_MIN_SNDBUF
2013-06-19 9:51 ` Eric Dumazet
@ 2013-06-19 9:57 ` Daniel Borkmann
2013-06-19 10:02 ` Eric Dumazet
2013-06-19 9:58 ` Eric Dumazet
1 sibling, 1 reply; 5+ messages in thread
From: Daniel Borkmann @ 2013-06-19 9:57 UTC (permalink / raw)
To: Eric Dumazet; +Cc: davem, netdev
On 06/19/2013 11:51 AM, Eric Dumazet wrote:
> On Wed, 2013-06-19 at 11:18 +0200, Daniel Borkmann wrote:
>> The current situation is that SOCK_MIN_RCVBUF is 2048 + sizeof(struct sk_buff))
>> while SOCK_MIN_SNDBUF is 2048. Since in both cases, skb->truesize is used for
>> sk_{r,w}mem_alloc accounting, we should have both sizes equal and adjusted
>> through the macro SKB_TRUESIZE(), which is also used elsewhere to adjust sk
>> buffer sizes. The minor adaption in sk_stream_moderate_sndbuf() is to silence
>> a warning by using a typed max macro, as similarly done in SOCK_MIN_RCVBUF
>> occurences, that would appear otherwise.
>>
>> Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
>> ---
>> include/net/sock.h | 11 ++++++-----
>> 1 file changed, 6 insertions(+), 5 deletions(-)
>>
>> diff --git a/include/net/sock.h b/include/net/sock.h
>> index ac8e181..189ef98 100644
>> --- a/include/net/sock.h
>> +++ b/include/net/sock.h
>> @@ -2045,18 +2045,19 @@ static inline void sk_wake_async(struct sock *sk, int how, int band)
>> sock_wake_async(sk->sk_socket, how, band);
>> }
>>
>> -#define SOCK_MIN_SNDBUF 2048
>> /*
>> - * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
>> - * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
>> + * Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
>> + * need sizeof(sk_buff) + sizeof(skb_shared_info) + MTU + padding, unless
>> + * net driver perform copybreak.
>> */
>> -#define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
>> +#define SOCK_MIN_RCVBUF SKB_TRUESIZE(2048)
>> +#define SOCK_MIN_SNDBUF SKB_TRUESIZE(2048)
>>
>>
>> static inline void sk_stream_moderate_sndbuf(struct sock *sk)
>> {
>> if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
>> sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
>> - sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
>> + sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
>> }
>> }
>>
>
> Funny you send this patch, because I prepared a similar patch
> yesterday ;)
Hehe, that is indeed funny. :-)
> My motivation was a bit different, because we hit a (small) regression
> here in Google for some applications setting low SO_SNDBUF/SO_RCVBUF
> values, because of new TCP needs :
>
> Minimal skb truesize in transmit path is indeed SKB_TRUESIZE(2048) after
> commit f07d960df33c5aef ("tcp: avoid frag allocation for small frames")
>
> And tcp sendmsg() tries to limit skb size to half the congestion window,
> meaning we try to build two skbs at minimum.
>
> So I believe that we need :
>
> /* TCP works better if we can build two skbs at minimum */
> #define SOCK_MIN_SNDBUF (2 * SKB_TRUESIZE(2048))
Ok, if you prefer, I can send an update.
Thanks,
Daniel
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [PATCH net-next] net: sock: adapt SOCK_MIN_RCVBUF and SOCK_MIN_SNDBUF
2013-06-19 9:57 ` Daniel Borkmann
@ 2013-06-19 10:02 ` Eric Dumazet
0 siblings, 0 replies; 5+ messages in thread
From: Eric Dumazet @ 2013-06-19 10:02 UTC (permalink / raw)
To: Daniel Borkmann; +Cc: davem, netdev
On Wed, 2013-06-19 at 11:57 +0200, Daniel Borkmann wrote:
> Ok, if you prefer, I can send an update.
>
Yes please.
By the way, I used exactly :
-#define SOCK_MIN_SNDBUF 2048
+/* minimum tcp skb truesize is 2048 + sizeof(struct sk_buff) */
+#define TCP_SKB_MIN_TRUESIZE (2048 + sizeof(struct sk_buff))
+
+#define SOCK_MIN_SNDBUF (2 * TCP_SKB_MIN_TRUESIZE)
because SKB_TRUESIZE(2048) adds
SKB_DATA_ALIGN(sizeof(struct skb_shared_info))
But in case of TCP skbs, the skb_shared_info is part of the 2048 bytes
allocation for skb->head
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH net-next] net: sock: adapt SOCK_MIN_RCVBUF and SOCK_MIN_SNDBUF
2013-06-19 9:51 ` Eric Dumazet
2013-06-19 9:57 ` Daniel Borkmann
@ 2013-06-19 9:58 ` Eric Dumazet
1 sibling, 0 replies; 5+ messages in thread
From: Eric Dumazet @ 2013-06-19 9:58 UTC (permalink / raw)
To: Daniel Borkmann; +Cc: davem, netdev
On Wed, 2013-06-19 at 02:51 -0700, Eric Dumazet wrote:
> On Wed, 2013-06-19 at 11:18 +0200, Daniel Borkmann wrote:
> > The current situation is that SOCK_MIN_RCVBUF is 2048 + sizeof(struct sk_buff))
> > while SOCK_MIN_SNDBUF is 2048. Since in both cases, skb->truesize is used for
> > sk_{r,w}mem_alloc accounting, we should have both sizes equal and adjusted
> > through the macro SKB_TRUESIZE(), which is also used elsewhere to adjust sk
> > buffer sizes. The minor adaption in sk_stream_moderate_sndbuf() is to silence
> > a warning by using a typed max macro, as similarly done in SOCK_MIN_RCVBUF
> > occurences, that would appear otherwise.
> >
> > Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
> > ---
> > include/net/sock.h | 11 ++++++-----
> > 1 file changed, 6 insertions(+), 5 deletions(-)
> >
> > diff --git a/include/net/sock.h b/include/net/sock.h
> > index ac8e181..189ef98 100644
> > --- a/include/net/sock.h
> > +++ b/include/net/sock.h
> > @@ -2045,18 +2045,19 @@ static inline void sk_wake_async(struct sock *sk, int how, int band)
> > sock_wake_async(sk->sk_socket, how, band);
> > }
> >
> > -#define SOCK_MIN_SNDBUF 2048
> > /*
> > - * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
> > - * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
> > + * Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
> > + * need sizeof(sk_buff) + sizeof(skb_shared_info) + MTU + padding, unless
> > + * net driver perform copybreak.
> > */
> > -#define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
> > +#define SOCK_MIN_RCVBUF SKB_TRUESIZE(2048)
> > +#define SOCK_MIN_SNDBUF SKB_TRUESIZE(2048)
> >
> >
> > static inline void sk_stream_moderate_sndbuf(struct sock *sk)
> > {
> > if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
> > sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
> > - sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
> > + sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
> > }
> > }
> >
>
> Funny you send this patch, because I prepared a similar patch
> yesterday ;)
>
> My motivation was a bit different, because we hit a (small) regression
> here in Google for some applications setting low SO_SNDBUF/SO_RCVBUF
> values, because of new TCP needs :
>
> Minimal skb truesize in transmit path is indeed SKB_TRUESIZE(2048) after
> commit f07d960df33c5aef ("tcp: avoid frag allocation for small frames")
>
> And tcp sendmsg() tries to limit skb size to half the congestion window,
> meaning we try to build two skbs at minimum.
>
> So I believe that we need :
>
> /* TCP works better if we can build two skbs at minimum */
> #define SOCK_MIN_SNDBUF (2 * SKB_TRUESIZE(2048))
Or more exactly :
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2013-06-19 10:02 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2013-06-19 9:18 [PATCH net-next] net: sock: adapt SOCK_MIN_RCVBUF and SOCK_MIN_SNDBUF Daniel Borkmann
2013-06-19 9:51 ` Eric Dumazet
2013-06-19 9:57 ` Daniel Borkmann
2013-06-19 10:02 ` Eric Dumazet
2013-06-19 9:58 ` Eric Dumazet
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.