Merge branch 'for-uring-ubufops' into HEAD

Pavel Begunkov says:

====================
implement io_uring notification (ubuf_info) stacking (net part)

To have per request buffer notifications each zerocopy io_uring send
request allocates a new ubuf_info. However, as an skb can carry only
one uarg, it may force the stack to create many small skbs hurting
performance in many ways.

The patchset implements notification, i.e. an io_uring's ubuf_info
extension, stacking. It attempts to link ubuf_info's into a list,
allowing to have multiple of them per skb.

liburing/examples/send-zerocopy shows up 6 times performance improvement
for TCP with 4KB bytes per send, and levels it with MSG_ZEROCOPY. Without
the patchset it requires much larger sends to utilise all potential.

bytes  | before | after (Kqps)
1200   | 195    | 1023
4000   | 193    | 1386
8000   | 154    | 1058
====================

Link: https://lore.kernel.org/all/cover.1713369317.git.asml.silence@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

9 files changed, 69 insertions, 36 deletions

diff --git a/drivers/net/tap.c b/drivers/net/tap.c
index 9f0495e8df4d..bfdd3875fe86 100644
--- a/drivers/net/tap.c
+++ b/drivers/net/tap.c
@@ -754,7 +754,7 @@ static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
 		skb_zcopy_init(skb, msg_control);
 	} else if (msg_control) {
 		struct ubuf_info *uarg = msg_control;
-		uarg->callback(NULL, uarg, false);
+		uarg->ops->complete(NULL, uarg, false);
 	}
 
 	dev_queue_xmit(skb);
diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 92da8c03d960..9254bca2813d 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1906,7 +1906,7 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 		skb_zcopy_init(skb, msg_control);
 	} else if (msg_control) {
 		struct ubuf_info *uarg = msg_control;
-		uarg->callback(NULL, uarg, false);
+		uarg->ops->complete(NULL, uarg, false);
 	}
 
 	skb_reset_network_header(skb);
diff --git a/drivers/net/xen-netback/common.h b/drivers/net/xen-netback/common.h
index 1fcbd83f7ff2..17421da139f2 100644
--- a/drivers/net/xen-netback/common.h
+++ b/drivers/net/xen-netback/common.h
@@ -390,9 +390,8 @@ bool xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb);
 
 void xenvif_carrier_on(struct xenvif *vif);
 
-/* Callback from stack when TX packet can be released */
-void xenvif_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *ubuf,
-			      bool zerocopy_success);
+/* Callbacks from stack when TX packet can be released */
+extern const struct ubuf_info_ops xenvif_ubuf_ops;
 
 static inline pending_ring_idx_t nr_pending_reqs(struct xenvif_queue *queue)
 {
diff --git a/drivers/net/xen-netback/interface.c b/drivers/net/xen-netback/interface.c
index 7cff90aa8d24..65db5f14465f 100644
--- a/drivers/net/xen-netback/interface.c
+++ b/drivers/net/xen-netback/interface.c
@@ -593,7 +593,7 @@ int xenvif_init_queue(struct xenvif_queue *queue)
 
 	for (i = 0; i < MAX_PENDING_REQS; i++) {
 		queue->pending_tx_info[i].callback_struct = (struct ubuf_info_msgzc)
-			{ { .callback = xenvif_zerocopy_callback },
+			{ { .ops = &xenvif_ubuf_ops },
 			  { { .ctx = NULL,
 			      .desc = i } } };
 		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c
index 48254fc07d64..5836995d6774 100644
--- a/drivers/net/xen-netback/netback.c
+++ b/drivers/net/xen-netback/netback.c
@@ -1157,7 +1157,7 @@ static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *s
 	uarg = skb_shinfo(skb)->destructor_arg;
 	/* increase inflight counter to offset decrement in callback */
 	atomic_inc(&queue->inflight_packets);
-	uarg->callback(NULL, uarg, true);
+	uarg->ops->complete(NULL, uarg, true);
 	skb_shinfo(skb)->destructor_arg = NULL;
 
 	/* Fill the skb with the new (local) frags. */
@@ -1279,8 +1279,9 @@ static int xenvif_tx_submit(struct xenvif_queue *queue)
 	return work_done;
 }
 
-void xenvif_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *ubuf_base,
-			      bool zerocopy_success)
+static void xenvif_zerocopy_callback(struct sk_buff *skb,
+				     struct ubuf_info *ubuf_base,
+				     bool zerocopy_success)
 {
 	unsigned long flags;
 	pending_ring_idx_t index;
@@ -1313,6 +1314,10 @@ void xenvif_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *ubuf_base,
 	xenvif_skb_zerocopy_complete(queue);
 }
 
+const struct ubuf_info_ops xenvif_ubuf_ops = {
+	.complete = xenvif_zerocopy_callback,
+};
+
 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
 {
 	struct gnttab_unmap_grant_ref *gop;
diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index c64ded183f8d..f16279351db5 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -380,7 +380,7 @@ static void vhost_zerocopy_signal_used(struct vhost_net *net,
 	}
 }
 
-static void vhost_zerocopy_callback(struct sk_buff *skb,
+static void vhost_zerocopy_complete(struct sk_buff *skb,
 				    struct ubuf_info *ubuf_base, bool success)
 {
 	struct ubuf_info_msgzc *ubuf = uarg_to_msgzc(ubuf_base);
@@ -408,6 +408,10 @@ static void vhost_zerocopy_callback(struct sk_buff *skb,
 	rcu_read_unlock_bh();
 }
 
+static const struct ubuf_info_ops vhost_ubuf_ops = {
+	.complete = vhost_zerocopy_complete,
+};
+
 static inline unsigned long busy_clock(void)
 {
 	return local_clock() >> 10;
@@ -879,7 +883,7 @@ static void handle_tx_zerocopy(struct vhost_net *net, struct socket *sock)
 			vq->heads[nvq->upend_idx].len = VHOST_DMA_IN_PROGRESS;
 			ubuf->ctx = nvq->ubufs;
 			ubuf->desc = nvq->upend_idx;
-			ubuf->ubuf.callback = vhost_zerocopy_callback;
+			ubuf->ubuf.ops = &vhost_ubuf_ops;
 			ubuf->ubuf.flags = SKBFL_ZEROCOPY_FRAG;
 			refcount_set(&ubuf->ubuf.refcnt, 1);
 			msg.msg_control = &ctl;
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 4072a7ee3859..f76825e5b92a 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -527,6 +527,13 @@ enum {
 #define SKBFL_ALL_ZEROCOPY	(SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \
 				 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS)
 
+struct ubuf_info_ops {
+	void (*complete)(struct sk_buff *, struct ubuf_info *,
+			 bool zerocopy_success);
+	/* has to be compatible with skb_zcopy_set() */
+	int (*link_skb)(struct sk_buff *skb, struct ubuf_info *uarg);
+};
+
 /*
  * 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.
@@ -536,8 +543,7 @@ enum {
  * The desc field is used to track userspace buffer index.
  */
 struct ubuf_info {
-	void (*callback)(struct sk_buff *, struct ubuf_info *,
-			 bool zerocopy_success);
+	const struct ubuf_info_ops *ops;
 	refcount_t refcnt;
 	u8 flags;
 };
@@ -1671,14 +1677,13 @@ static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset)
 }
 #endif
 
+extern const struct ubuf_info_ops msg_zerocopy_ubuf_ops;
+
 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
 				       struct ubuf_info *uarg);
 
 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
 
-void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
-			   bool success);
-
 int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk,
 			    struct sk_buff *skb, struct iov_iter *from,
 			    size_t length);
@@ -1766,13 +1771,13 @@ static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
 static inline void net_zcopy_put(struct ubuf_info *uarg)
 {
 	if (uarg)
-		uarg->callback(NULL, uarg, true);
+		uarg->ops->complete(NULL, uarg, true);
 }
 
 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref)
 {
 	if (uarg) {
-		if (uarg->callback == msg_zerocopy_callback)
+		if (uarg->ops == &msg_zerocopy_ubuf_ops)
 			msg_zerocopy_put_abort(uarg, have_uref);
 		else if (have_uref)
 			net_zcopy_put(uarg);
@@ -1786,7 +1791,7 @@ static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success)
 
 	if (uarg) {
 		if (!skb_zcopy_is_nouarg(skb))
-			uarg->callback(skb, uarg, zerocopy_success);
+			uarg->ops->complete(skb, uarg, zerocopy_success);
 
 		skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY;
 	}
diff --git a/io_uring/notif.c b/io_uring/notif.c
index d3e703c37aba..204f0fc7d966 100644
--- a/io_uring/notif.c
+++ b/io_uring/notif.c
@@ -24,7 +24,7 @@ static void io_notif_complete_tw_ext(struct io_kiocb *notif, struct io_tw_state
 	io_req_task_complete(notif, ts);
 }
 
-static void io_tx_ubuf_callback(struct sk_buff *skb, struct ubuf_info *uarg,
+static void io_tx_ubuf_complete(struct sk_buff *skb, struct ubuf_info *uarg,
 				bool success)
 {
 	struct io_notif_data *nd = container_of(uarg, struct io_notif_data, uarg);
@@ -45,19 +45,27 @@ static void io_tx_ubuf_callback_ext(struct sk_buff *skb, struct ubuf_info *uarg,
 		else if (!success && !nd->zc_copied)
 			WRITE_ONCE(nd->zc_copied, true);
 	}
-	io_tx_ubuf_callback(skb, uarg, success);
+	io_tx_ubuf_complete(skb, uarg, success);
 }
 
+static const struct ubuf_info_ops io_ubuf_ops = {
+	.complete = io_tx_ubuf_complete,
+};
+
+static const struct ubuf_info_ops io_ubuf_ops_ext = {
+	.complete = io_tx_ubuf_callback_ext,
+};
+
 void io_notif_set_extended(struct io_kiocb *notif)
 {
 	struct io_notif_data *nd = io_notif_to_data(notif);
 
-	if (nd->uarg.callback != io_tx_ubuf_callback_ext) {
+	if (nd->uarg.ops != &io_ubuf_ops_ext) {
 		nd->account_pages = 0;
 		nd->zc_report = false;
 		nd->zc_used = false;
 		nd->zc_copied = false;
-		nd->uarg.callback = io_tx_ubuf_callback_ext;
+		nd->uarg.ops = &io_ubuf_ops_ext;
 		notif->io_task_work.func = io_notif_complete_tw_ext;
 	}
 }
@@ -80,7 +88,7 @@ struct io_kiocb *io_alloc_notif(struct io_ring_ctx *ctx)
 
 	nd = io_notif_to_data(notif);
 	nd->uarg.flags = IO_NOTIF_UBUF_FLAGS;
-	nd->uarg.callback = io_tx_ubuf_callback;
+	nd->uarg.ops = &io_ubuf_ops;
 	refcount_set(&nd->uarg.refcnt, 1);
 	return notif;
 }
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 37c858dc11a6..0c8b82750000 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -1652,7 +1652,7 @@ static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
 		return NULL;
 	}
 
-	uarg->ubuf.callback = msg_zerocopy_callback;
+	uarg->ubuf.ops = &msg_zerocopy_ubuf_ops;
 	uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
 	uarg->len = 1;
 	uarg->bytelen = size;
@@ -1678,7 +1678,7 @@ struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
 		u32 bytelen, next;
 
 		/* there might be non MSG_ZEROCOPY users */
-		if (uarg->callback != msg_zerocopy_callback)
+		if (uarg->ops != &msg_zerocopy_ubuf_ops)
 			return NULL;
 
 		/* realloc only when socket is locked (TCP, UDP cork),
@@ -1789,8 +1789,8 @@ release:
 	sock_put(sk);
 }
 
-void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
-			   bool success)
+static void msg_zerocopy_complete(struct sk_buff *skb, struct ubuf_info *uarg,
+				  bool success)
 {
 	struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg);
 
@@ -1799,7 +1799,6 @@ void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
 	if (refcount_dec_and_test(&uarg->refcnt))
 		__msg_zerocopy_callback(uarg_zc);
 }
-EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
 
 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
 {
@@ -1809,10 +1808,15 @@ void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
 	uarg_to_msgzc(uarg)->len--;
 
 	if (have_uref)
-		msg_zerocopy_callback(NULL, uarg, true);
+		msg_zerocopy_complete(NULL, uarg, true);
 }
 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
 
+const struct ubuf_info_ops msg_zerocopy_ubuf_ops = {
+	.complete = msg_zerocopy_complete,
+};
+EXPORT_SYMBOL_GPL(msg_zerocopy_ubuf_ops);
+
 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
 			     struct msghdr *msg, int len,
 			     struct ubuf_info *uarg)
@@ -1820,11 +1824,18 @@ int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
 	struct ubuf_info *orig_uarg = skb_zcopy(skb);
 	int err, orig_len = skb->len;
 
-	/* An skb can only point to one uarg. This edge case happens when
-	 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
-	 */
-	if (orig_uarg && uarg != orig_uarg)
-		return -EEXIST;
+	if (uarg->ops->link_skb) {
+		err = uarg->ops->link_skb(skb, uarg);
+		if (err)
+			return err;
+	} else {
+		/* An skb can only point to one uarg. This edge case happens
+		 * when TCP appends to an skb, but zerocopy_realloc triggered
+		 * a new alloc.
+		 */
+		if (orig_uarg && uarg != orig_uarg)
+			return -EEXIST;
+	}
 
 	err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len);
 	if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
@@ -1838,7 +1849,8 @@ int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
 		return err;
 	}
 
-	skb_zcopy_set(skb, uarg, NULL);
+	if (!uarg->ops->link_skb)
+		skb_zcopy_set(skb, uarg, NULL);
 	return skb->len - orig_len;
 }
 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);