Re: [RFC] skb: avoid unnecessary reallocations in __skb_cow

On Tue, 2012-05-29 at 14:41 +0200, Felix Fietkau wrote:

On 2012-05-29 2:34 PM, Eric Dumazet wrote:

quoted

On Sun, 2012-05-27 at 17:26 +0200, Felix Fietkau wrote:

quoted

At the beginning of __skb_cow, headroom gets set to a minimum of
NET_SKB_PAD. This causes unnecessary reallocations if the buffer was not
cloned and the headroom is just below NET_SKB_PAD, but still more than the
amount requested by the caller.
This was showing up frequently in my tests on VLAN tx, where
vlan_insert_tag calls skb_cow_head(skb, VLAN_HLEN).

Fix this by only setting the headroom delta if either there is less
headroom than specified by the caller, or if reallocation has to be done
anyway because the skb was cloned.

Signed-off-by: Felix Fietkau <redacted>
---
 include/linux/skbuff.h |    9 ++++++---
 1 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 0e50171..1898471 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h

@@ -1894,12 +1894,15 @@ static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len
 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
 			    int cloned)
 {
+	unsigned int alloc_headroom = headroom;
 	int delta = 0;
 
 	if (headroom < NET_SKB_PAD)
-		headroom = NET_SKB_PAD;
-	if (headroom > skb_headroom(skb))
-		delta = headroom - skb_headroom(skb);
+		alloc_headroom = NET_SKB_PAD;
+	if (headroom > skb_headroom(skb) ||
+	    (cloned && alloc_headroom > skb_headroom(skb))) {
+		delta = alloc_headroom - skb_headroom(skb);
+	}
 
 	if (delta || cloned)
 		return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,

Nice catch.

Scratching my head on this one. Why not the obvious fix ?

If we're reallocating anyway, we might as well put in more headroom than
requested, in case something else needs even more than that.

Locally generated packets should have enough headroom, and for forward
paths, we already have NET_SKB_PAD bytes of headroom.

Adding yet another NET_SKB_PAD extra space is overkill, unless you have
a real use case in mind ?