Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Pavel Begunkov <asml.silence@gmail.com>
To:     io-uring@vger.kernel.org, netdev@vger.kernel.org
Cc:     Jens Axboe <axboe@kernel.dk>,
        "David S . Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>, kernel-team@fb.com,
        linux-kernel@vger.kernel.org, xen-devel@lists.xenproject.org,
        Wei Liu <wei.liu@kernel.org>, Paul Durrant <paul@xen.org>,
        kvm@vger.kernel.org, virtualization@lists.linux-foundation.org,
        "Michael S . Tsirkin" <mst@redhat.com>,
        Jason Wang <jasowang@redhat.com>,
        Pavel Begunkov <asml.silence@gmail.com>
Subject: [RFC net-next io_uring 07/11] net/tcp: optimise tcp ubuf refcounting
Date:   Wed, 10 Aug 2022 16:49:15 +0100
Message-Id: <9a42187cdc9ce034fd23179c7b31d7cc6a54bd45.1660124059.git.asml.silence@gmail.com>
In-Reply-To: <cover.1660124059.git.asml.silence@gmail.com>
References: <cover.1660124059.git.asml.silence@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Add UARGFL_CALLER_PINNED letting protocols know that the caller holds a
reference to the ubuf_info and so it doesn't need additional refcounting
for purposes of keeping it alive. With that TCP can save a refcount
put/get pair per send when used with ->msg_ubuf.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 include/linux/skbuff.h | 7 +++++++
 net/ipv4/tcp.c         | 9 ++++++---
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 2b2e0020030b..45fe7f0648d0 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -522,6 +522,13 @@ enum {
 #define SKBFL_ALL_ZEROCOPY	(SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
 				 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
 
+enum {
+	/* The caller holds a reference during the submission so the ubuf won't
+	 * be freed until we return.
+	 */
+	UARGFL_CALLER_PINNED = BIT(0),
+};
+
 /*
  * The callback notifies userspace to release buffers when skb DMA is done in
  * lower device, the skb last reference should be 0 when calling this.
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 3152da8f4763..4925107de57d 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -1229,7 +1229,8 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 
 		if (msg->msg_ubuf) {
 			uarg = msg->msg_ubuf;
-			net_zcopy_get(uarg);
+			if (!(uarg->flags & UARGFL_CALLER_PINNED))
+				net_zcopy_get(uarg);
 			zc = sk->sk_route_caps & NETIF_F_SG;
 		} else if (sock_flag(sk, SOCK_ZEROCOPY)) {
 			uarg = msg_zerocopy_realloc(sk, size, skb_zcopy(skb));
@@ -1455,7 +1456,8 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 		tcp_push(sk, flags, mss_now, tp->nonagle, size_goal);
 	}
 out_nopush:
-	net_zcopy_put(uarg);
+	if (uarg && !(uarg->flags & UARGFL_CALLER_PINNED))
+		net_zcopy_put(uarg);
 	return copied + copied_syn;
 
 do_error:
@@ -1464,7 +1466,8 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 	if (copied + copied_syn)
 		goto out;
 out_err:
-	net_zcopy_put_abort(uarg, true);
+	if (uarg && !(uarg->flags & UARGFL_CALLER_PINNED))
+		net_zcopy_put_abort(uarg, true);
 	err = sk_stream_error(sk, flags, err);
 	/* make sure we wake any epoll edge trigger waiter */
 	if (unlikely(tcp_rtx_and_write_queues_empty(sk) && err == -EAGAIN)) {
-- 
2.37.0