Re: [PATCH v3] tcp: add support for SO_PEEK_OFF

[Jon Maloy <jmaloy@redhat.com>]

On 2024-02-15 12:46, Eric Dumazet wrote:
> On Thu, Feb 15, 2024 at 6:41 PM Paolo Abeni <pabeni@redhat.com> wrote:
>> Note: please send text-only email to netdev.
>>
>> On Thu, 2024-02-15 at 10:11 -0500, Jon Maloy wrote:
>>> I wonder if the following could be acceptable:
>>>
>>>   if (flags & MSG_PEEK)
>>>          sk_peek_offset_fwd(sk, used);
>>>   else if (peek_offset > 0)
>>>         sk_peek_offset_bwd(sk, used);
>>>
>>>   peek_offset is already present in the data cache, and if it has the value
>>>   zero it means either that that sk->sk_peek_off is unused (-1) or actually is zero.
>>>   Either way, no rewind is needed in that case.
>> I agree the above should avoid touching cold cachelines in the
>> fastpath, and looks functionally correct to me.
>>
>> The last word is up to Eric :)
>>
> An actual patch seems needed.
>
> In the current form, local variable peek_offset is 0 when !MSG_PEEK.
>
> So the "else if (peek_offset > 0)" would always be false.
>
Yes, of course. This wouldn't work unless we read sk->sk_peek_off at the 
beginning of the function.
I will look at the other suggestions.

///jon