[TCP]: Account skb overhead in tcp_fragment

[TCP]: Account skb overhead in tcp_fragment

[Herbert Xu]

[-- Attachment #1: Type: text/plain, Size: 536 bytes --]

Hi Dave:

Since sk_stream_alloc_pskb takes an extra argument that accounts for
paged data all we need to do to account sk_buff overhead correctly
is to use that instead of sk_stream_alloc_skb.

This patch does just that for both tcp_fragment and tso_fragment.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

[-- Attachment #2: tcp-fragment-charge.patch --]
[-- Type: text/plain, Size: 1062 bytes --]

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 44df1db..fb40d32 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -548,12 +548,10 @@
 		return -ENOMEM;
 
 	/* Get a new skb... force flag on. */
-	buff = sk_stream_alloc_skb(sk, nsize, GFP_ATOMIC);
+	buff = sk_stream_alloc_pskb(sk, nsize, skb->len - len - nsize, GFP_ATOMIC);
 	if (buff == NULL)
 		return -ENOMEM; /* We'll just try again later. */
-
-	buff->truesize = skb->len - len;
-	skb->truesize -= buff->truesize;
+	sk_charge_skb(sk, buff);
 
 	/* Correct the sequence numbers. */
 	TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len;
@@ -1035,12 +1033,10 @@
 	if (skb->len != skb->data_len)
 		return tcp_fragment(sk, skb, len, mss_now);
 
-	buff = sk_stream_alloc_pskb(sk, 0, 0, GFP_ATOMIC);
+	buff = sk_stream_alloc_pskb(sk, 0, nlen, GFP_ATOMIC);
 	if (unlikely(buff == NULL))
 		return -ENOMEM;
-
-	buff->truesize = nlen;
-	skb->truesize -= nlen;
+	sk_charge_skb(sk, buff);
 
 	/* Correct the sequence numbers. */
 	TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len;

Re: [TCP]: Account skb overhead in tcp_fragment

[Herbert Xu]

[-- Attachment #1: Type: text/plain, Size: 628 bytes --]

On Thu, Apr 20, 2006 at 11:01:11AM +1000, herbert wrote:
> Hi Dave:
> 
> Since sk_stream_alloc_pskb takes an extra argument that accounts for
> paged data all we need to do to account sk_buff overhead correctly
> is to use that instead of sk_stream_alloc_skb.

Here is a better version that does not double-count the paged bits
that get moved across.
 
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

[-- Attachment #2: tcp-fragment-charge-2.patch --]
[-- Type: text/plain, Size: 890 bytes --]

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 44df1db..a28ae59 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -533,6 +533,7 @@
 	struct tcp_sock *tp = tcp_sk(sk);
 	struct sk_buff *buff;
 	int nsize, old_factor;
+	int nlen;
 	u16 flags;
 
 	BUG_ON(len > skb->len);
@@ -552,8 +553,10 @@
 	if (buff == NULL)
 		return -ENOMEM; /* We'll just try again later. */
 
-	buff->truesize = skb->len - len;
-	skb->truesize -= buff->truesize;
+	sk_charge_skb(sk, buff);
+	nlen = skb->len - len - nsize;
+	buff->truesize += nlen;
+	skb->truesize -= nlen;
 
 	/* Correct the sequence numbers. */
 	TCP_SKB_CB(buff)->seq = TCP_SKB_CB(skb)->seq + len;
@@ -1039,7 +1042,8 @@
 	if (unlikely(buff == NULL))
 		return -ENOMEM;
 
-	buff->truesize = nlen;
+	sk_charge_skb(sk, buff);
+	buff->truesize += nlen;
 	skb->truesize -= nlen;
 
 	/* Correct the sequence numbers. */

Re: [TCP]: Account skb overhead in tcp_fragment

[David S. Miller]

From: Herbert Xu <herbert@gondor.apana.org.au>
Date: Thu, 20 Apr 2006 13:26:02 +1000

> On Thu, Apr 20, 2006 at 11:01:11AM +1000, herbert wrote:
> > Hi Dave:
> > 
> > Since sk_stream_alloc_pskb takes an extra argument that accounts for
> > paged data all we need to do to account sk_buff overhead correctly
> > is to use that instead of sk_stream_alloc_skb.
> 
> Here is a better version that does not double-count the paged bits
> that get moved across.
>  
> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Looks good, I'll apply this.