On 9/1/20 6:52 PM, Ramamurthy, Harshitha wrote:

quoted

On 8/31/20 1:25 PM, Harshitha Ramamurthy wrote:

quoted

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h index

a613750d5515..bffe93b526e7 100644

--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h

@@ -3576,6 +3576,14 @@ union bpf_attr {
  * 		the data in *dst*. This is a wrapper of copy_from_user().
  * 	Return
  * 		0 on success, or a negative error in case of failure.
+ *
+ * u32 bpf_get_xdp_hash(struct xdp_buff *xdp_md)

I thought there was a change recently making the uapi reference xdp_md;
xdp_buff is not exported as part of the uapi.

Not sure what you mean - other xdp related helper functions still use xdp_buff as an argument. Could you point me to an example of what you are referring to?

a patch from Jakub that is apparently not committed yet.

quoted

quoted

+	memset(&keys, 0, sizeof(keys));
+	__skb_flow_dissect(dev_net(xdp->rxq->dev), NULL,

&flow_keys_dissector,

quoted

+			   &keys, data, eth->h_proto, sizeof(*eth), len,
+			   FLOW_DISSECTOR_F_STOP_AT_FLOW_LABEL);

By STOP_AT_FLOW_LABEL I take it you want this to be an L3 hash. Why not
add a flags argument to the helper and let the hash be L3 or L4?

I wrote this exactly how skb_get_hash calls skb_flow_dissect - with the same flag STOP_AT_FLOW_LABEL.  So it should already cover L3 and L4 hash, right? From what I understand STOP_AT_FLOW_LABEL flag is used to only stop parsing when a flow label is seen in ipv6 packets. 

right; missed that. That means this new helper will always do an L4 hash
for tcp/udp. Adding a flags argument now for the uapi allows a later
change to make it an L3 hash.