Re: [RFC PATCH bpf-next v6 2/3] net: Add additional bit to support clockid_t timestamp type

[Willem de Bruijn <willemdebruijn.kernel@gmail.com>]

Martin KaFai Lau wrote:
> On 5/3/24 8:13 PM, Abhishek Chauhan wrote:
> > diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
> > index fe86cadfa85b..c3d852eecb01 100644
> > --- a/net/ipv4/ip_output.c
> > +++ b/net/ipv4/ip_output.c
> > @@ -1457,7 +1457,10 @@ 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;
> > +	if (sk_is_tcp(sk))
> 
> This seems not catching all IPPROTO_TCP case. In particular, the percpu 
> "ipv4_tcp_sk" is SOCK_RAW. sk_is_tcp() is checking SOCK_STREAM:
> 
> void __init tcp_v4_init(void)
> {
> 
> 	/* ... */
> 	res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
> 				   IPPROTO_TCP, &init_net);
> 
> 	/* ... */
> }
> 
> "while :; do ./test_progs -t tc_redirect/tc_redirect_dtime || break; done" 
> failed pretty often exactly in this case.
> 

Interesting. The TCP stack opens non TCP sockets.

Initializing sk->sk_clockid for this socket should address that.