Re: [PATCH] net: tcp_drop adds `reason` parameter for tracing

From: Jakub Kicinski <kuba@kernel.org>
To: Zhongya Yan <yan2228598786@gmail.com>
Cc: netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
	edumazet@google.com, rostedt@goodmis.org, mingo@redhat.com,
	davem@davemloft.net, yoshfuji@linux-ipv6.org, dsahern@kernel.org,
	hengqi.chen@gmail.com, yhs@fb.com
Subject: Re: [PATCH] net: tcp_drop adds `reason` parameter for tracing
Date: Tue, 24 Aug 2021 07:20:49 -0700	[thread overview]
Message-ID: <20210824072049.76789bba@kicinski-fedora-pc1c0hjn.dhcp.thefacebook.com> (raw)
In-Reply-To: <20210824125140.190253-1-yan2228598786@gmail.com>

On Tue, 24 Aug 2021 05:51:40 -0700 Zhongya Yan wrote:
> +enum tcp_drop_reason {
> +	TCP_OFO_QUEUE = 1,
> +	TCP_DATA_QUEUE_OFO = 2,
> +	TCP_DATA_QUEUE = 3,
> +	TCP_PRUNE_OFO_QUEUE = 4,
> +	TCP_VALIDATE_INCOMING = 5,
> +	TCP_RCV_ESTABLISHED = 6,
> +	TCP_RCV_SYNSENT_STATE_PROCESS = 7,
> +	TCP_RCV_STATE_PROCESS = 8
> +};

This is basically tracking the caller, each may have multiple reasons
for dropping. Is tracking the caller sufficient? Should we at least
make this a bitmask so we can set multiple bits (caller and more
precise reason)? Or are we going to add another field in that case?

> -static void tcp_drop(struct sock *sk, struct sk_buff *skb)
> +static void __tcp_drop(struct sock *sk,
> +		   struct sk_buff *skb)
>  {
>  	sk_drops_add(sk, skb);
>  	__kfree_skb(skb);
>  }

Why keep this function if there is only one caller?

> +/* tcp_drop whit reason,for epbf trace
> + */

This comment is (a) misspelled, (b) doesn't add much value.

> +static void tcp_drop(struct sock *sk, struct sk_buff *skb,
> +		 enum tcp_drop_reason reason)
> +{
> +	trace_tcp_drop(sk, skb, reason);
> +	__tcp_drop(sk, skb);
> +}