Re: [xdp-hints] Re: [PATCH bpf-next v4 08/15] veth: Support RX XDP metadata

Jesper Dangaard Brouer [off-list ref] writes:

On 13/12/2022 21.42, Stanislav Fomichev wrote:

quoted

On Tue, Dec 13, 2022 at 7:55 AM Jesper Dangaard Brouer
[off-list ref] wrote:

quoted

On 13/12/2022 03.35, Stanislav Fomichev wrote:

quoted

The goal is to enable end-to-end testing of the metadata for AF_XDP.

Cc: John Fastabend <john.fastabend@gmail.com>
Cc: David Ahern <redacted>
Cc: Martin KaFai Lau <martin.lau@linux.dev>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Willem de Bruijn <willemb@google.com>
Cc: Jesper Dangaard Brouer <redacted>
Cc: Anatoly Burakov <redacted>
Cc: Alexander Lobakin <redacted>
Cc: Magnus Karlsson <redacted>
Cc: Maryam Tahhan <redacted>
Cc: xdp-hints@xdp-project.net
Cc: netdev@vger.kernel.org
Signed-off-by: Stanislav Fomichev <redacted>
---
   drivers/net/veth.c | 24 ++++++++++++++++++++++++
   1 file changed, 24 insertions(+)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index 04ffd8cb2945..d5491e7a2798 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c

@@ -118,6 +118,7 @@ static struct {

   struct veth_xdp_buff {
       struct xdp_buff xdp;
+     struct sk_buff *skb;
   };

   static int veth_get_link_ksettings(struct net_device *dev,

@@ -602,6 +603,7 @@ static struct xdp_frame *veth_xdp_rcv_one(struct veth_rq *rq,

               xdp_convert_frame_to_buff(frame, xdp);
               xdp->rxq = &rq->xdp_rxq;
+             vxbuf.skb = NULL;

               act = bpf_prog_run_xdp(xdp_prog, xdp);

@@ -823,6 +825,7 @@ static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq,
       __skb_push(skb, skb->data - skb_mac_header(skb));
       if (veth_convert_skb_to_xdp_buff(rq, xdp, &skb))
               goto drop;
+     vxbuf.skb = skb;

       orig_data = xdp->data;
       orig_data_end = xdp->data_end;

@@ -1601,6 +1604,21 @@ static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp)
       }
   }

+static int veth_xdp_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp)
+{
+     *timestamp = ktime_get_mono_fast_ns();

This should be reading the hardware timestamp in the SKB.

Details: This hardware timestamp in the SKB is located in
skb_shared_info area, which is also available for xdp_frame (currently
used for multi-buffer purposes).  Thus, when adding xdp-hints "store"
functionality, it would be natural to store the HW TS in the same place.
Making the veth skb/xdp_frame code paths able to share code.

Does something like the following look acceptable as well?

*timestamp = skb_hwtstamps(_ctx->skb)->hwtstamp;
if (!*timestamp)
         *timestamp = ktime_get_mono_fast_ns(); /* sw fallback */

How can the BPF programmer tell the difference between getting hardware 
or software timestamp?

This will get really confusing when someone implements a tcpdump feature
(like/extend xdpdump) and some packets (e.g. PTP packets) have HW
timestamps and some don't.  The time sequence in the pcap will be strange.

quoted

Because I'd like to be able to test this path in the selftests. As
long as I get some number from veth_xdp_rx_timestamp, I can test it.
No amount of SOF_TIMESTAMPING_{SOFTWARE,RX_SOFTWARE,RAW_HARDWARE}
triggers non-zero hwtstamp for xsk receive path. Any suggestions?

You could implement the "store" operation I mentioned before.
For testing you can store an arbitrary value in the timestamp and check
it later by reading it back.

I can see you have changed the API to send down a pointer. Thus, a
simple flag could implement the writing the provided timestamp.

Regarding flags for reading the timestamp.  Should we be able to specify
what clock type we are asking for?
Have you notice that tcpdump can ask for different types of
timestamps[1]. e.g. for hardware timestamps it is either
'adapter_unsynced' or 'adaptor'. (See semantic in [1])

  # tcpdump -ni eth1 -j adapter_unsynced --time-stamp-precision=nano

I don't think it makes sense for XDP to *ask* for a specific timestamp
(any individual packet probably only has one, right?). But we could add
a way to query which type of timestamp is available, either as a second
argument to the same function, or as a separate one. Same thing for the
hash, BTW (skb_set_hash() also takes a type argument).

I guess the easiest would be to just add a second parameter to both
getter functions for the type, but maybe there's a performance argument
for having it be a separate kfunc (at least for timestamp)?

-Toke