On Thu, 24 Mar 2022 09:50:06 +0100 Paolo Abeni wrote:

quoted

+/**
+ * DOC: dataref and headerless skbs
+ *
+ * Transport layers send out clones of data skbs they hold for retransmissions.
+ * To allow lower layers of the stack to prepend their headers
+ * we split &skb_shared_info.dataref into two halves.
+ * The lower 16 bits count the overall number of references.
+ * The higher 16 bits indicate number of data-only references.
+ * skb_header_cloned() checks if skb is allowed to add / write the headers.  

Thank you very much for the IMHO much needed documentation!

Please allow me to do some non-native-english-speaker biased comments;)

The previous patch uses the form  "payload data" instead of data-only,
I think it would be clearer using consistently one or the other. I
personally would go for "payload-data-only" (which is probably a good
reason to pick a different option).

That starts to get long. Let me go for payload everywhere.

quoted

- * All users must obey the rule that the skb->data reference count must be
- * greater than or equal to the payload reference count.
+ * The creator of the skb (e.g. TCP) marks its data-only skb as &sk_buff.nohdr
+ * (via __skb_header_release()). Any clone created from marked skb will get
+ * &sk_buff.hdr_len populated with the available headroom.
+ * If it's the only clone in existence it's able to modify the headroom at will.  

I think it would be great if we explicitly list the expected sequence,
e.g.
	<alloc skb>
	skb_reserve
	__skb_header_release
	skb_clone

Will do!