splice: do not checksum AF_UNIX sockets

[PATCH v2 net] splice: do not checksum AF_UNIX sockets

Posted by Frederik Deweerdt 2 weeks, 2 days ago

When `skb_splice_from_iter` was introduced, it inadvertently added
checksumming for AF_UNIX sockets. This resulted in significant
slowdowns, for example when using sendfile over unix sockets.

Using the test code in [1] in my test setup (2G single core qemu),
the client receives a 1000M file in:
- without the patch: 1482ms (+/- 36ms)
- with the patch: 652.5ms (+/- 22.9ms)

This commit addresses the issue by marking checksumming as unnecessary in
`unix_stream_sendmsg`

Cc: stable@vger.kernel.org
Signed-off-by: Frederik Deweerdt <deweerdt.lkml@gmail.com>
Fixes: 2e910b95329c ("net: Add a function to splice pages into an skbuff for MSG_SPLICE_PAGES")
---
 net/unix/af_unix.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index 001ccc55ef0f..6b1762300443 100644
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -2313,6 +2313,7 @@ static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
 		fds_sent = true;
 
 		if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
+			skb->ip_summed = CHECKSUM_UNNECESSARY;
 			err = skb_splice_from_iter(skb, &msg->msg_iter, size,
 						   sk->sk_allocation);
 			if (err < 0) {
-- 
2.44.1

Re: [PATCH v2 net] splice: do not checksum AF_UNIX sockets

Posted by Joe Damato 2 weeks, 2 days ago

On Mon, Dec 09, 2024 at 09:06:48PM -0800, Frederik Deweerdt wrote:
> When `skb_splice_from_iter` was introduced, it inadvertently added
> checksumming for AF_UNIX sockets. This resulted in significant
> slowdowns, for example when using sendfile over unix sockets.
> 
> Using the test code in [1] in my test setup (2G single core qemu),
> the client receives a 1000M file in:
> - without the patch: 1482ms (+/- 36ms)
> - with the patch: 652.5ms (+/- 22.9ms)
> 
> This commit addresses the issue by marking checksumming as unnecessary in
> `unix_stream_sendmsg`
> 
> Cc: stable@vger.kernel.org
> Signed-off-by: Frederik Deweerdt <deweerdt.lkml@gmail.com>
> Fixes: 2e910b95329c ("net: Add a function to splice pages into an skbuff for MSG_SPLICE_PAGES")
> ---
>  net/unix/af_unix.c | 1 +
>  1 file changed, 1 insertion(+)

Reviewed-by: Joe Damato <jdamato@fastly.com>

Re: [PATCH v2 net] splice: do not checksum AF_UNIX sockets

Posted by Eric Dumazet 2 weeks, 2 days ago

On Tue, Dec 10, 2024 at 6:06 AM Frederik Deweerdt
<deweerdt.lkml@gmail.com> wrote:
>
> When `skb_splice_from_iter` was introduced, it inadvertently added
> checksumming for AF_UNIX sockets. This resulted in significant
> slowdowns, for example when using sendfile over unix sockets.
>
> Using the test code in [1] in my test setup (2G single core qemu),
> the client receives a 1000M file in:
> - without the patch: 1482ms (+/- 36ms)
> - with the patch: 652.5ms (+/- 22.9ms)
>
> This commit addresses the issue by marking checksumming as unnecessary in
> `unix_stream_sendmsg`
>
> Cc: stable@vger.kernel.org
> Signed-off-by: Frederik Deweerdt <deweerdt.lkml@gmail.com>
> Fixes: 2e910b95329c ("net: Add a function to splice pages into an skbuff for MSG_SPLICE_PAGES")
> ---

Reviewed-by: Eric Dumazet <edumazet@google.com>

Re: [PATCH v2 net] splice: do not checksum AF_UNIX sockets

Posted by Kuniyuki Iwashima 2 weeks, 2 days ago

From: Frederik Deweerdt <deweerdt.lkml@gmail.com>
Date: Mon, 9 Dec 2024 21:06:48 -0800
> When `skb_splice_from_iter` was introduced, it inadvertently added
> checksumming for AF_UNIX sockets. This resulted in significant
> slowdowns, for example when using sendfile over unix sockets.
> 
> Using the test code in [1] in my test setup (2G single core qemu),
> the client receives a 1000M file in:
> - without the patch: 1482ms (+/- 36ms)
> - with the patch: 652.5ms (+/- 22.9ms)
> 
> This commit addresses the issue by marking checksumming as unnecessary in
> `unix_stream_sendmsg`
> 
> Cc: stable@vger.kernel.org
> Signed-off-by: Frederik Deweerdt <deweerdt.lkml@gmail.com>
> Fixes: 2e910b95329c ("net: Add a function to splice pages into an skbuff for MSG_SPLICE_PAGES")
> ---

For the future submission, it would be nice to explain changes
between versions and add the old patch link under '---' here.

The patch itself looks good to me.

Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>


>  net/unix/af_unix.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
> index 001ccc55ef0f..6b1762300443 100644
> --- a/net/unix/af_unix.c
> +++ b/net/unix/af_unix.c
> @@ -2313,6 +2313,7 @@ static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
>  		fds_sent = true;
>  
>  		if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
> +			skb->ip_summed = CHECKSUM_UNNECESSARY;
>  			err = skb_splice_from_iter(skb, &msg->msg_iter, size,
>  						   sk->sk_allocation);
>  			if (err < 0) {
> -- 
> 2.44.1

Re: [PATCH v2 net] splice: do not checksum AF_UNIX sockets

Posted by Frederik Deweerdt 2 weeks, 2 days ago

On Tue, Dec 10, 2024 at 05:17:21PM +0900, Kuniyuki Iwashima wrote:
> From: Frederik Deweerdt <deweerdt.lkml@gmail.com>
> Date: Mon, 9 Dec 2024 21:06:48 -0800
> > When `skb_splice_from_iter` was introduced, it inadvertently added
> > checksumming for AF_UNIX sockets. This resulted in significant
> > slowdowns, for example when using sendfile over unix sockets.
> > 
> > Using the test code in [1] in my test setup (2G single core qemu),
> > the client receives a 1000M file in:
> > - without the patch: 1482ms (+/- 36ms)
> > - with the patch: 652.5ms (+/- 22.9ms)
> > 
> > This commit addresses the issue by marking checksumming as unnecessary in
> > `unix_stream_sendmsg`
> > 
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Frederik Deweerdt <deweerdt.lkml@gmail.com>
> > Fixes: 2e910b95329c ("net: Add a function to splice pages into an skbuff for MSG_SPLICE_PAGES")
> > ---
> 
> For the future submission, it would be nice to explain changes
> between versions and add the old patch link under '---' here.
> 
Will do, thank you.

> The patch itself looks good to me.
> 
> Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
> 

Thanks!
Frederik
> >  net/unix/af_unix.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
> > index 001ccc55ef0f..6b1762300443 100644
> > --- a/net/unix/af_unix.c
> > +++ b/net/unix/af_unix.c
> > @@ -2313,6 +2313,7 @@ static int unix_stream_sendmsg(struct socket *sock, struct msghdr *msg,
> >  		fds_sent = true;
> >  
> >  		if (unlikely(msg->msg_flags & MSG_SPLICE_PAGES)) {
> > +			skb->ip_summed = CHECKSUM_UNNECESSARY;
> >  			err = skb_splice_from_iter(skb, &msg->msg_iter, size,
> >  						   sk->sk_allocation);
> >  			if (err < 0) {
> > -- 
> > 2.44.1