Re: [EXT] [PATCH net v2] ipv4: ip_gre: Avoid skb_pull() failure in ipgre_xmit()

[Shigeru Yoshida]

Hi Suman,

On Sun, 3 Dec 2023 06:58:19 +0000, Suman Ghosh wrote:
> Hi Shigeru,
> 
>>diff --git a/net/ipv4/ip_gre.c b/net/ipv4/ip_gre.c index
>>22a26d1d29a0..5169c3c72cff 100644
>>--- a/net/ipv4/ip_gre.c
>>+++ b/net/ipv4/ip_gre.c
>>@@ -635,15 +635,18 @@ static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
>> 	}
>>
>> 	if (dev->header_ops) {
>>+		int pull_len = tunnel->hlen + sizeof(struct iphdr);
>>+
>> 		if (skb_cow_head(skb, 0))
>> 			goto free_skb;
>>
>> 		tnl_params = (const struct iphdr *)skb->data;
>>
>>-		/* Pull skb since ip_tunnel_xmit() needs skb->data pointing
>>-		 * to gre header.
>>-		 */
>>-		skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
>>+		if (!pskb_network_may_pull(skb, pull_len))
> [Suman] Since this is transmit path, should we add unlikely() here?

Thanks for your comment.

I traced this function and found that pskb_may_pull_reason() seems to
have appropriate likely() and unlikely() as Eric says.

I'm new to Linux networking. Could you kindly explain the background
of your suggestion?

I understand that a transmit path must be as fast as possible, so we
should use unlikely() for rare cases such like this error path. Am I
correct?

Thanks,
Shigeru

>>+			goto free_skb;
>>+
>>+		/* ip_tunnel_xmit() needs skb->data pointing to gre header. */
>>+		skb_pull(skb, pull_len);
>> 		skb_reset_mac_header(skb);
>>
>> 		if (skb->ip_summed == CHECKSUM_PARTIAL &&
>>--
>>2.41.0
>>
>