[RFC net-next io_uring 08/11] net: let callers provide ->msg_ubuf refs

[Pavel Begunkov]

Some msg_ubuf providers like io_uring can keep elaborated ubuf_info
reference batching and caching, so it will be of benefit to let the
network layer to optionally steal some of the cached refs.

Add UARGFL_GIFT_REF, if set the caller has at least one extra reference
that it can gift away. If the network decides to take the ref it should
clear the flag.

Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
---
 include/linux/skbuff.h | 14 ++++++++++++++
 net/ipv4/ip_output.c   |  1 +
 net/ipv6/ip6_output.c  |  1 +
 3 files changed, 16 insertions(+)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 45fe7f0648d0..972ec676e222 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -527,6 +527,11 @@ enum {
 	 * be freed until we return.
 	 */
 	UARGFL_CALLER_PINNED = BIT(0),
+
+	/* The caller can gift one ubuf reference. The flag should be cleared
+	 * when the reference is taken.
+	 */
+	UARGFL_GIFT_REF = BIT(1),
 };
 
 /*
@@ -1709,6 +1714,15 @@ static inline void net_zcopy_put(struct ubuf_info *uarg)
 		uarg->callback(NULL, uarg, true);
 }
 
+static inline bool net_zcopy_get_gift_ref(struct ubuf_info *uarg)
+{
+	bool has_ref;
+
+	has_ref = uarg->flags & UARGFL_GIFT_REF;
+	uarg->flags &= ~UARGFL_GIFT_REF;
+	return has_ref;
+}
+
 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
 {
 	if (uarg) {
diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 546897a4b4fa..9d42b6dd6b78 100644
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1032,6 +1032,7 @@ static int __ip_append_data(struct sock *sk,
 				paged = true;
 				zc = true;
 				uarg = msg->msg_ubuf;
+				extra_uref = net_zcopy_get_gift_ref(uarg);
 			}
 		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
 			uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index 6d4f01a0cf6e..8d8a8bbdb8df 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -1557,6 +1557,7 @@ static int __ip6_append_data(struct sock *sk,
 				paged = true;
 				zc = true;
 				uarg = msg->msg_ubuf;
+				extra_uref = net_zcopy_get_gift_ref(uarg);
 			}
 		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
 			uarg = msg_zerocopy_realloc(sk, length, skb_zcopy(skb));
-- 
2.37.0