Re: [RFC PATCH bpf-next v5 2/2] net: Add additional bit to support clockid_t timestamp type

[Abhishek Chauhan (ABC)]

On 4/30/2024 1:26 PM, Willem de Bruijn wrote:
> Abhishek Chauhan (ABC) wrote:
>>
>>
>> On 4/26/2024 4:50 PM, Martin KaFai Lau wrote:
>>> On 4/26/24 11:46 AM, Abhishek Chauhan (ABC) wrote:
>>>>>> diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
>>>>>> index 591226dcde26..f195b31d6e75 100644
>>>>>> --- a/net/ipv4/ip_output.c
>>>>>> +++ b/net/ipv4/ip_output.c
>>>>>> @@ -1457,7 +1457,7 @@ struct sk_buff *__ip_make_skb(struct sock *sk,
>>>>>>          skb->priority = (cork->tos != -1) ? cork->priority: READ_ONCE(sk->sk_priority);
>>>>>>        skb->mark = cork->mark;
>>>>>> -    skb->tstamp = cork->transmit_time;
>>>>>> +    skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);
>>>>> hmm... I think this will break for tcp. This sequence in particular:
> 
> Good catch, thanks!
> 
>>>>>
>>>>> tcp_v4_timewait_ack()
>>>>>    tcp_v4_send_ack()
>>>>>      ip_send_unicast_reply()
>>>>>        ip_push_pending_frames()
>>>>>          ip_finish_skb()
>>>>>            __ip_make_skb()
>>>>>              /* sk_clockid is REAL but cork->transmit_time should be in mono */
>>>>>              skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);;
>>>>>
>>>>> I think I hit it from time to time when running the test in this patch set.
>>>>>
>>>> do you think i need to check for protocol type here . since tcp uses Mono and the rest according to the new design is based on
>>>> sk->sk_clockid
>>>> if (iph->protocol == IPPROTO_TCP)
>>>>     skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, CLOCK_MONOTONIC);
>>>> else
>>>>     skb_set_tstamp_type_frm_clkid(skb, cork->transmit_time, sk->sk_clockid);
>>>
>>> Looks ok. iph->protocol is from sk->sk_protocol. I would defer to Willem input here.
>>>
>>> There is at least one more place that needs this protocol check, __ip6_make_skb().
>>
>> Sounds good. I will wait for Willem to comment here. 
> 
> This would be sk_is_tcp(sk).
> 
> I think we want to avoid special casing if we can. Note the if.
> 
> If TCP always uses monotonic, we could consider initializing
> sk_clockid to CLOCK_MONONOTIC in tcp_init_sock.
> 
> I guess TCP logic currently entirely ignores sk_clockid. If we are to
> start using this, then setsocktop SO_TXTIME must explicitly fail or
> ignore for TCP sockets, or silently skip the write.
> 
> All of that is more complexity. Than is maybe warranted for this one
> case. So no objections from me to special casing using sk_is_tcp(sk)
> either.
> 
>

Thanks Willem and Martin for all the support and reviews. 
I will take care of this in the next RFC patch
- adding the sk_is_tcp check and setting tstamp_type to mono for tcp

> 
>