Re: [PATCH net-next v4 2/4] net: tap: track dropped skb via kfree_skb_reason()

From: Dongli Zhang <dongli.zhang@oracle.com>
To: Jakub Kicinski <kuba@kernel.org>
Cc: netdev@vger.kernel.org, bpf@vger.kernel.org,
	linux-kernel@vger.kernel.org, davem@davemloft.net,
	rostedt@goodmis.org, mingo@redhat.com, ast@kernel.org,
	daniel@iogearbox.net, andrii@kernel.org, imagedong@tencent.com,
	joao.m.martins@oracle.com, joe.jin@oracle.com, dsahern@gmail.com,
	edumazet@google.com
Subject: Re: [PATCH net-next v4 2/4] net: tap: track dropped skb via kfree_skb_reason()
Date: Wed, 2 Mar 2022 13:44:20 -0800	[thread overview]
Message-ID: <92322369-abb6-be0e-ab52-1d1ebbe38a9c@oracle.com> (raw)
In-Reply-To: <20220302110300.1ac78804@kicinski-fedora-PC1C0HJN.hsd1.ca.comcast.net>

Hi Jakub,

On 3/2/22 11:03 AM, Jakub Kicinski wrote:
> On Wed, 2 Mar 2022 09:43:29 -0800 Dongli Zhang wrote:
>> On 3/1/22 6:42 PM, Jakub Kicinski wrote:
>>> On Sat, 26 Feb 2022 00:49:27 -0800 Dongli Zhang wrote:  
>>>> +	SKB_DROP_REASON_SKB_CSUM,	/* sk_buff checksum error */  
>>>
>>> Can we spell it out a little more? It sounds like the checksum was
>>> incorrect. Will it be clear that computing the checksum failed, rather
>>> than checksum validation failed?  
>>
>> I am just trying to make the reasons as generic as possible so that:
>>
>> 1. We may minimize the number of reasons.
>>
>> 2. People may re-use the same reason for all CSUM related issue.
> 
> The generic nature is fine, my concern is to clearly differentiate
> errors in _validating_ the checksum from errors in _generating_ them.
> "sk_buff checksum error" does not explain which one had taken place.

This is for skb_checksum_help() and it is for csum computation.

Therefore, I will keep SKB_DROP_REASON_SKB_CSUM and add 'computation' or
'generating' to the comments.

> 
>>>> +	SKB_DROP_REASON_SKB_COPY_DATA,	/* failed to copy data from or to
>>>> +					 * sk_buff
>>>> +					 */  
>>>
>>> Here should we specify that it's copying from user space?  
>>
>> Same as above. I am minimizing the number of reasons so that any memory copy for
>> sk_buff may re-use this reason.
> 
> IIUC this failure is equivalent to user passing an invalid buffer. 
> I mean something like:
> 
> 	send(fd, (void *)random(), 1000, 0);
> 
> I'd be tempted to call the reason something link SKB_UCOPY_FAULT.
> To indicate it's a problem copying from user space. EFAULT is the
> typical errno for that. WDYT?
> 

So far the reason is used for below functions' return value:

- tap_get_user() -> zerocopy_sg_from_iter()
- tap_get_user() -> skb_copy_datagram_from_iter()
- tun_net_xmit() -> skb_orphan_frags_rx() -> skb_copy_ubufs()

I will switch to SKB_UCOPY_FAULT.

Thank you very much!

Dongli Zhang