On Wed, Oct 9, 2024 at 2:53 AM Willem de Bruijn
<willemdebruijn.kernel@xxxxxxxxx> wrote:
>
> Jason Xing wrote:
> > From: Jason Xing <kernelxing@xxxxxxxxxxx>
> >
> > Once we set BPF_SOCK_OPS_TX_TIMESTAMP_OPT_CB_FLAG flag here, there
> > are three points in the previous patches where generating timestamps
> > works. Let us make the basic bpf mechanism for timestamping feature
> > work finally.
> >
> > We can use like this as a simple example in bpf program:
> > __section("sockops")
> >
> > case BPF_SOCK_OPS_TX_TIMESTAMP_OPT_CB:
> > dport = bpf_ntohl(skops->remote_port);
> > sport = skops->local_port;
> > skops->reply = SOF_TIMESTAMPING_TX_SCHED;
> > bpf_sock_ops_cb_flags_set(skops, BPF_SOCK_OPS_TX_TIMESTAMP_OPT_CB_FLAG);
> > case BPF_SOCK_OPS_TS_SCHED_OPT_CB:
> > bpf_printk(...);
> >
> > Signed-off-by: Jason Xing <kernelxing@xxxxxxxxxxx>
>
> > /* List of TCP states. There is a build check in net/ipv4/tcp.c to detect
> > diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> > index 82cc4a5633ce..ddf4089779b5 100644
> > --- a/net/ipv4/tcp.c
> > +++ b/net/ipv4/tcp.c
> > @@ -477,12 +477,37 @@ void tcp_init_sock(struct sock *sk)
> > }
> > EXPORT_SYMBOL(tcp_init_sock);
> >
> > +static u32 bpf_tcp_tx_timestamp(struct sock *sk)
> > +{
> > + u32 flags;
> > +
> > + flags = tcp_call_bpf(sk, BPF_SOCK_OPS_TX_TS_OPT_CB, 0, NULL);
> > + if (flags <= 0)
> > + return 0;
> > +
> > + if (flags & ~SOF_TIMESTAMPING_MASK)
> > + return 0;
> > +
> > + if (!(flags & SOF_TIMESTAMPING_TX_RECORD_MASK))
> > + return 0;
> > +
> > + return flags;
> > +}
> > +
> > static void tcp_tx_timestamp(struct sock *sk, struct sockcm_cookie *sockc)
> > {
> > struct sk_buff *skb = tcp_write_queue_tail(sk);
> > u32 tsflags = sockc->tsflags;
> > + u32 flags;
> > +
> > + if (!skb)
> > + return;
> > +
> > + flags = bpf_tcp_tx_timestamp(sk);
> > + if (flags)
> > + tsflags = flags;
>
> So this feature overwrites the flags set by the user?
It only overrides each last skb instead of the whole socket so that
some time if we don't want to use this bpf program any more, we could
easily and directly detach it without having to find a proper time to
clear the fields in struct sock. That's the advantage of setting
through each sendmsg call, compared to bpf_setsockopt method.
> Ideally we would use an entirely separate field for BPF admin
> timestamping requests.
I understand what you mean. I'm not that familiar with how a bpf
extension actually implements, so I dug into how RTO min time can be
affected by bpf programs (see BPF_SOCK_OPS_TIMEOUT_INIT as an
example). It also modifies the existing field.
Thanks,
Jason
