[PATCH net-next] be2net: dont pull too much data in skb linear part

[PATCH net-next] be2net: dont pull too much data in skb linear part

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

skb_fill_rx_data() pulls 64 byte of data in skb->data

Its too much for TCP (with no options) on IPv4, as total size of headers
is 14 + 40 = 54

This means tcp stack and splice() are suboptimal, since tcp payload
is in part in tcp->data, and in part in skb frag.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 drivers/net/ethernet/emulex/benet/be_main.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/emulex/benet/be_main.c b/drivers/net/ethernet/emulex/benet/be_main.c
index 7e989d0..f18375c 100644
--- a/drivers/net/ethernet/emulex/benet/be_main.c
+++ b/drivers/net/ethernet/emulex/benet/be_main.c
@@ -1228,16 +1228,16 @@ static void skb_fill_rx_data(struct be_rx_obj *rxo, struct sk_buff *skb,
 	/* Copy data in the first descriptor of this completion */
 	curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
 
-	/* Copy the header portion into skb_data */
-	hdr_len = min(BE_HDR_LEN, curr_frag_len);
-	memcpy(skb->data, start, hdr_len);
 	skb->len = curr_frag_len;
 	if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
+		memcpy(skb->data, start, curr_frag_len);
 		/* Complete packet has now been moved to data */
 		put_page(page_info->page);
 		skb->data_len = 0;
 		skb->tail += curr_frag_len;
 	} else {
+		hdr_len = ETH_HLEN;
+		memcpy(skb->data, start, hdr_len);
 		skb_shinfo(skb)->nr_frags = 1;
 		skb_frag_set_page(skb, 0, page_info->page);
 		skb_shinfo(skb)->frags[0].page_offset =

RE: [PATCH net-next] be2net: dont pull too much data in skb linear part

[Padmanabh.Ratnakar]

> -----Original Message-----
> From: Eric Dumazet [mailto:eric.dumazet@gmail.com]
> Sent: Friday, July 13, 2012 6:50 PM
> To: Ratnakar, Padmanabh
> Cc: netdev
> Subject: [PATCH net-next] be2net: dont pull too much data in skb linear part
> 
> From: Eric Dumazet <edumazet@google.com>
> 
> skb_fill_rx_data() pulls 64 byte of data in skb->data
> 
> Its too much for TCP (with no options) on IPv4, as total size of headers is 14 +
> 40 = 54
> 
> This means tcp stack and splice() are suboptimal, since tcp payload is in part in
> tcp->data, and in part in skb frag.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
>  drivers/net/ethernet/emulex/benet/be_main.c |    6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/ethernet/emulex/benet/be_main.c
> b/drivers/net/ethernet/emulex/benet/be_main.c
> index 7e989d0..f18375c 100644
> --- a/drivers/net/ethernet/emulex/benet/be_main.c
> +++ b/drivers/net/ethernet/emulex/benet/be_main.c
> @@ -1228,16 +1228,16 @@ static void skb_fill_rx_data(struct be_rx_obj
> *rxo, struct sk_buff *skb,
>  	/* Copy data in the first descriptor of this completion */
>  	curr_frag_len = min(rxcp->pkt_size, rx_frag_size);
> 
> -	/* Copy the header portion into skb_data */
> -	hdr_len = min(BE_HDR_LEN, curr_frag_len);
> -	memcpy(skb->data, start, hdr_len);
>  	skb->len = curr_frag_len;
>  	if (curr_frag_len <= BE_HDR_LEN) { /* tiny packet */
> +		memcpy(skb->data, start, curr_frag_len);
>  		/* Complete packet has now been moved to data */
>  		put_page(page_info->page);
>  		skb->data_len = 0;
>  		skb->tail += curr_frag_len;
>  	} else {
> +		hdr_len = ETH_HLEN;
> +		memcpy(skb->data, start, hdr_len);
>  		skb_shinfo(skb)->nr_frags = 1;
>  		skb_frag_set_page(skb, 0, page_info->page);
>  		skb_shinfo(skb)->frags[0].page_offset =
> 
Change looks good to me. I tested this and it is working.
Thanks
Acked-by: Padmanabh Ratnakar <padmanabh.ratnakar@emulex.com>

Re: [PATCH net-next] be2net: dont pull too much data in skb linear part

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Fri, 13 Jul 2012 15:19:41 +0200

> From: Eric Dumazet <edumazet@google.com>
> 
> skb_fill_rx_data() pulls 64 byte of data in skb->data
> 
> Its too much for TCP (with no options) on IPv4, as total size of headers
> is 14 + 40 = 54
> 
> This means tcp stack and splice() are suboptimal, since tcp payload
> is in part in tcp->data, and in part in skb frag.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied.