Re: [RFC PATCH bpf-next v4 1/2] net: Rename mono_delivery_time to tstamp_type for scalabilty

[Willem de Bruijn]

Abhishek Chauhan (ABC) wrote:
> 
> 
> On 4/18/2024 1:11 PM, Willem de Bruijn wrote:
> > Abhishek Chauhan (ABC) wrote:
> >>
> >>
> >> On 4/18/2024 11:47 AM, Willem de Bruijn wrote:
> >>> Abhishek Chauhan wrote:
> >>>> mono_delivery_time was added to check if skb->tstamp has delivery
> >>>> time in mono clock base (i.e. EDT) otherwise skb->tstamp has
> >>>> timestamp in ingress and delivery_time at egress.
> >>>>
> >>>> Renaming the bitfield from mono_delivery_time to tstamp_type is for
> >>>> extensibilty for other timestamps such as userspace timestamp
> >>>> (i.e. SO_TXTIME) set via sock opts.
> >>>>
> >>>> As we are renaming the mono_delivery_time to tstamp_type, it makes
> >>>> sense to start assigning tstamp_type based on enum defined
> >>>> in this commit.
> >>>>
> >>>> Earlier we used bool arg flag to check if the tstamp is mono in
> >>>> function skb_set_delivery_time, Now the signature of the functions
> >>>> accepts tstamp_type to distinguish between mono and real time.
> >>>>
> >>>> In future tstamp_type:1 can be extended to support userspace timestamp
> >>>> by increasing the bitfield.
> >>>>
> >>>> Link: https://lore.kernel.org/netdev/bc037db4-58bb-4861-ac31-a361a93841d3@xxxxxxxxx/
> >>>> Signed-off-by: Abhishek Chauhan <quic_abchauha@xxxxxxxxxxx>
> >>>
> >>>> +/**
> >>>> + * tstamp_type:1 can take 2 values each
> >>>> + * represented by time base in skb
> >>>> + * 0x0 => real timestamp_type
> >>>> + * 0x1 => mono timestamp_type
> >>>> + */
> >>>> +enum skb_tstamp_type {
> >>>> +	SKB_CLOCK_REAL,	/* Time base is skb is REALTIME */
> >>>> +	SKB_CLOCK_MONO,	/* Time base is skb is MONOTONIC */
> >>>> +};
> >>>> +
> >>>
> >>> Can drop the comments. These names are self documenting.
> >>
> >> Noted! . I will take care of this
> >>>
> >>>>  /**
> >>>>   * DOC: Basic sk_buff geometry
> >>>>   *
> >>>> @@ -819,7 +830,7 @@ typedef unsigned char *sk_buff_data_t;
> >>>>   *	@dst_pending_confirm: need to confirm neighbour
> >>>>   *	@decrypted: Decrypted SKB
> >>>>   *	@slow_gro: state present at GRO time, slower prepare step required
> >>>> - *	@mono_delivery_time: When set, skb->tstamp has the
> >>>> + *	@tstamp_type: When set, skb->tstamp has the
> >>>>   *		delivery_time in mono clock base (i.e. EDT).  Otherwise, the
> >>>>   *		skb->tstamp has the (rcv) timestamp at ingress and
> >>>>   *		delivery_time at egress.
> >>>
> >>> Is this still correct? I think all egress does now annotate correctly
> >>> as SKB_CLOCK_MONO. So when not set it always is SKB_CLOCK_REAL.
> >>>
> >> That is correct. 
> >>
> >>>> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> >>>> index 61119d42b0fd..a062f88c47c3 100644
> >>>> --- a/net/ipv4/tcp_output.c
> >>>> +++ b/net/ipv4/tcp_output.c
> >>>> @@ -1300,7 +1300,7 @@ static int __tcp_transmit_skb(struct sock *sk, struct sk_buff *skb,
> >>>>  	tp = tcp_sk(sk);
> >>>>  	prior_wstamp = tp->tcp_wstamp_ns;
> >>>>  	tp->tcp_wstamp_ns = max(tp->tcp_wstamp_ns, tp->tcp_clock_cache);
> >>>> -	skb_set_delivery_time(skb, tp->tcp_wstamp_ns, true);
> >>>> +	skb_set_delivery_time(skb, tp->tcp_wstamp_ns, CLOCK_MONOTONIC);
> >>>
> >>> Multiple references to CLOCK_MONOTONIC left
> >>>
> >> I think i took care of all the references. Apologies if i didn't understand your comment here. 
> > 
> > On closer read, there is a type issue here.
> > 
> > skb_set_delivery_time takes a u8 tstamp_type. But it is often passed
> > a clockid_t, and that is also what the switch expects.
> > 
> > But it does also get called with a tstamp_type in code like the
> > following:
> > 
> > +       u8 tstamp_type = skb->tstamp_type;
> >         unsigned int hlen, ll_rs, mtu;
> >         ktime_t tstamp = skb->tstamp;
> >         struct ip_frag_state state;
> > @@ -82,7 +82,7 @@ static int nf_br_ip_fragment(struct net *net, struct sock *sk,
> >                         if (iter.frag)
> >                                 ip_fraglist_prepare(skb, &iter);
> >   
> > -                       skb_set_delivery_time(skb, tstamp, mono_delivery_time);
> > +                       skb_set_delivery_time(skb, tstamp, tstamp_type);
> > 
> > So maybe we need two variants, one that takes a tstamp_type and one
> > that tames a clockid_t?
> > 
> > The first can be simple, not switch needed. Just apply the two stores.
> I agree to what you are saying but clockid_t => points to int itself. 
> 
> For example :- 
> 		void qdisc_watchdog_init_clockid(struct qdisc_watchdog *wd, struct Qdisc *qdisc,
> 				 clockid_t clockid)
> 
> 		qdisc_watchdog_init_clockid(wd, qdisc, CLOCK_MONOTONIC); => sch_api.c
> 	       qdisc_watchdog_init_clockid(&q->watchdog, sch, q->clockid); =>sch_etf.c (q->clockid is int)

My concern is more that we use CLOCK_MONOTONIC and SKB_CLOCK_MONO
(and other clocks) interchangeably, without invariant checks to make
sure that they map onto the same integer value.