Re: [PATCH] sctp: do trace_sctp_probe after SACK validation and check

[Kevin Kou]

On 2019/12/23 21:26, Marcelo Ricardo Leitner wrote:
> On Sun, Dec 22, 2019 at 12:22:24PM +0800, Kevin Kou wrote:
> > On 2019/12/21 0:17, Marcelo Ricardo Leitner wrote:
> > > On Fri, Dec 20, 2019 at 04:47:03AM +0000, Kevin Kou wrote:
> > > > The function sctp_sf_eat_sack_6_2 now performs
> > > > the Verification Tag validation, Chunk length validation, Bogu check,
> > > > and also the detection of out-of-order SACK based on the RFC2960
> > > > Section 6.2 at the beginning, and finally performs the further
> > > > processing of SACK. The trace_sctp_probe now triggered before
> > > > the above necessary validation and check.
> > > >
> > > > This patch is to do the trace_sctp_probe after the necessary check
> > > > and validation to SACK.
> > > >
> > > > Signed-off-by: Kevin Kou <qdkevin.kou@xxxxxxxxx>
> > > > ---
> > > >    net/sctp/sm_statefuns.c | 3 ++-
> > > >    1 file changed, 2 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/net/sctp/sm_statefuns.c b/net/sctp/sm_statefuns.c
> > > > index 42558fa..b4a54df 100644
> > > > --- a/net/sctp/sm_statefuns.c
> > > > +++ b/net/sctp/sm_statefuns.c
> > > > @@ -3281,7 +3281,6 @@ enum sctp_disposition sctp_sf_eat_sack_6_2(struct net *net,
> > > >    	struct sctp_sackhdr *sackh;
> > > >    	__u32 ctsn;
> > > > -	trace_sctp_probe(ep, asoc, chunk);
> > > >    	if (!sctp_vtag_verify(chunk, asoc))
> > > >    		return sctp_sf_pdiscard(net, ep, asoc, type, arg, commands);
> > > > @@ -3319,6 +3318,8 @@ enum sctp_disposition sctp_sf_eat_sack_6_2(struct net *net,
> > > >    	if (!TSN_lt(ctsn, asoc->next_tsn))
> > > >    		return sctp_sf_violation_ctsn(net, ep, asoc, type, arg, commands);
> > > > +	trace_sctp_probe(ep, asoc, chunk);
> > > > +
> > >
> > > Moving it here will be after the check against ctsn_ack_point, which
> > > could cause duplicated SACKs to be missed from the log.
> >
> >
> > As this SCTP trace used to trace the changes of SCTP association state in
> > response to incoming packets(SACK). It is used for debugging SCTP congestion
> > control algorithms, so according to the code in include/trace/events/sctp.h,
> > the trace event mainly focus on congestion related information, and there is
> > no SACK Chunk related information printed. So it is hard to point out
> > whether the SACK is duplicate one or not based on this trace event.
>
> I see. Yet, it's quite odd to do debugging of congestion control
> algorithms without knowing how many TSNs/bytes are being acked by this
> ack, but let's keep that aside for now.
>
> I still can't agree with filtering out based the out-of-order SACK check
> (the TSN_lt(ctsn, asoc->ctsn_ack_point) check. That is valuable to
> congestion control debugging, because it will likely mean that the
> sender is working with fewer acks than it would like/expect.
>
> If you need to filter out them and have a "clean" list of what got in,
> then the fix it needs lies in adding support for logging the ctsn in
> the trace point itself (similarly to the pr_debug in there) and filter
> it on post-processing of the logs.
>
> I don't know how much of UAPI cover probe points. Hopefully we can add
> that information without having to create new probe points.
Thanks for your comments, In order to avoid affecting the UAPI, let's 
keep the existing print fields at present.

> PS: You can invert the check in
>          if (!TSN_lt(ctsn, asoc->next_tsn))
> to
>          if (TSN_lte(asoc->next_tsn, ctsn))
> and move it above, so it is done before the out-of-order check, and
> the trace point in between them.

I will make this change and commit another patch.

> > include/trace/events/sctp.h
> > 1. TRACE_EVENT(sctp_probe,
> >
> > TP_printk("asoc=%#llx mark=%#x bind_port=%d peer_port=%d pathmtu=%d "
> > 		  "rwnd=%u unack_data=%d",
> > 		  __entry->asoc, __entry->mark, __entry->bind_port,
> > 		  __entry->peer_port, __entry->pathmtu, __entry->rwnd,
> > 		  __entry->unack_data)
> >
> > 2. TRACE_EVENT(sctp_probe_path,
> >
> > TP_printk("asoc=%#llx%s ipaddr=%pISpc state=%u cwnd=%u ssthresh=%u "
> > 		  "flight_size=%u partial_bytes_acked=%u pathmtu=%u",
> > 		  __entry->asoc, __entry->primary ? "(*)" : "",
> > 		  __entry->ipaddr, __entry->state, __entry->cwnd,
> > 		  __entry->ssthresh, __entry->flight_size,
> > 		  __entry->partial_bytes_acked, __entry->pathmtu)
> >
> > > Yes, from the sender-side CC we don't care about it (yet), but it
> > > helps to spot probably avoidable retransmissions.
> > >
> > > I think this is cleaning up the noise too much. I can agree with
> > > moving it to after the chunk sanity tests, though.
> > >
> > > >    	/* Return this SACK for further processing.  */
> > > >    	sctp_add_cmd_sf(commands, SCTP_CMD_PROCESS_SACK, SCTP_CHUNK(chunk));
> > > > --
> > > > 1.8.3.1