netdev - Re: [PATCH net-next 2/2] vhost-net: reduce one userspace copy when building XDP buff

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <684b89de38e3_dcc452944e@willemb.c.googlers.com.notmuch>
Date: Thu, 12 Jun 2025 22:15:58 -0400
From: Willem de Bruijn <willemdebruijn.kernel@...il.com>
To: Jason Wang <jasowang@...hat.com>, 
 mst@...hat.com, 
 jasowang@...hat.com
Cc: eperezma@...hat.com, 
 kvm@...r.kernel.org, 
 virtualization@...ts.linux.dev, 
 netdev@...r.kernel.org, 
 willemdebruijn.kernel@...il.com, 
 davem@...emloft.net, 
 andrew+netdev@...n.ch, 
 edumazet@...gle.com, 
 kuba@...nel.org, 
 pabeni@...hat.com
Subject: Re: [PATCH net-next 2/2] vhost-net: reduce one userspace copy when
 building XDP buff

Jason Wang wrote:
> We used to do twice copy_from_iter() to copy virtio-net and packet
> separately. This introduce overheads for userspace access hardening as
> well as SMAP (for x86 it's stac/clac). So this patch tries to use one
> copy_from_iter() to copy them once and move the virtio-net header
> afterwards to reduce overheads.
> 
> Testpmd + vhost_net shows 10% improvement from 5.45Mpps to 6.0Mpps.
> 
> Signed-off-by: Jason Wang <jasowang@...hat.com>

Acked-by: Willem de Bruijn <willemb@...gle.com>

> ---
>  drivers/vhost/net.c | 13 ++++---------
>  1 file changed, 4 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
> index 777eb6193985..2845e0a473ea 100644
> --- a/drivers/vhost/net.c
> +++ b/drivers/vhost/net.c
> @@ -690,13 +690,13 @@ static int vhost_net_build_xdp(struct vhost_net_virtqueue *nvq,
>  	if (unlikely(!buf))
>  		return -ENOMEM;
>  
> -	copied = copy_from_iter(buf, sock_hlen, from);
> -	if (copied != sock_hlen) {
> +	copied = copy_from_iter(buf + pad - sock_hlen, len, from);
> +	if (copied != len) {
>  		ret = -EFAULT;
>  		goto err;
>  	}
>  
> -	gso = buf;
> +	gso = buf + pad - sock_hlen;
>  
>  	if (!sock_hlen)
>  		memset(buf, 0, pad);
> @@ -715,12 +715,7 @@ static int vhost_net_build_xdp(struct vhost_net_virtqueue *nvq,
>  		}
>  	}
>  
> -	len -= sock_hlen;
> -	copied = copy_from_iter(buf + pad, len, from);
> -	if (copied != len) {
> -		ret = -EFAULT;
> -		goto err;
> -	}
> +	memcpy(buf, buf + pad - sock_hlen, sock_hlen);

It's not trivial to see that the dst and src do not overlap, and does
does not need memmove.

Minimal pad that I can find is 32B and and maximal sock_hlen is 12B.

So this is safe. But not obviously so. Unfortunately, these offsets
are not all known at compile time, so a BUILD_BUG_ON is not possible.

>  	xdp_init_buff(xdp, buflen, NULL);
>  	xdp_prepare_buff(xdp, buf, pad, len, true);
> -- 
> 2.34.1
>