Merge branch 'tun-skb_array'

Jason Wang says:

====================
switch to use tx skb array in tun

This series tries to switch to use skb array in tun. This is used to
eliminate the spinlock contention between producer and consumer. The
conversion was straightforward: just introdce a tx skb array and use
it instead of sk_receive_queue.

A minor issue is to keep the tx_queue_len behaviour, since tun used to
use it for the length of sk_receive_queue. This is done through:

- add the ability to resize multiple rings at once to avoid handling
  partial resize failure for mutiple rings.
- add the support for zero length ring.
- introduce a notifier which was triggered when tx_queue_len was
  changed for a netdev.
- resize all queues during the tx_queue_len changing.

Tests shows about 15% improvement on guest rx pps:

Before: ~1300000pps
After : ~1500000pps

Changes from V3:
- fix kbuild warnings
- call NETDEV_CHANGE_TX_QUEUE_LEN on IFLA_TXQLEN

Changes from V2:
- add multiple rings resizing support for ptr_ring/skb_array
- add zero length ring support
- introdce a NETDEV_CHANGE_TX_QUEUE_LEN
- drop new flags

Changes from V1:
- switch to use skb array instead of a customized circular buffer
- add non-blocking support
- rename .peek to .peek_len
- drop lockless peeking since test show very minor improvement
====================

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-from-altitude: 34697 feet.
Signed-off-by: David S. Miller <davem@davemloft.net>

9 files changed, 255 insertions, 27 deletions

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 4884802..7475215 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -71,6 +71,7 @@
 #include <net/sock.h>
 #include <linux/seq_file.h>
 #include <linux/uio.h>
+#include <linux/skb_array.h>
 
 #include <asm/uaccess.h>
 
@@ -167,6 +168,7 @@ struct tun_file {
 	};
 	struct list_head next;
 	struct tun_struct *detached;
+	struct skb_array tx_array;
 };
 
 struct tun_flow_entry {
@@ -515,7 +517,11 @@ static struct tun_struct *tun_enable_queue(struct tun_file *tfile)
 
 static void tun_queue_purge(struct tun_file *tfile)
 {
-	skb_queue_purge(&tfile->sk.sk_receive_queue);
+	struct sk_buff *skb;
+
+	while ((skb = skb_array_consume(&tfile->tx_array)) != NULL)
+		kfree_skb(skb);
+
 	skb_queue_purge(&tfile->sk.sk_error_queue);
 }
 
@@ -560,6 +566,8 @@ static void __tun_detach(struct tun_file *tfile, bool clean)
 			    tun->dev->reg_state == NETREG_REGISTERED)
 				unregister_netdevice(tun->dev);
 		}
+		if (tun)
+			skb_array_cleanup(&tfile->tx_array);
 		sock_put(&tfile->sk);
 	}
 }
@@ -613,6 +621,7 @@ static void tun_detach_all(struct net_device *dev)
 static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filter)
 {
 	struct tun_file *tfile = file->private_data;
+	struct net_device *dev = tun->dev;
 	int err;
 
 	err = security_tun_dev_attach(tfile->socket.sk, tun->security);
@@ -642,6 +651,13 @@ static int tun_attach(struct tun_struct *tun, struct file *file, bool skip_filte
 		if (!err)
 			goto out;
 	}
+
+	if (!tfile->detached &&
+	    skb_array_init(&tfile->tx_array, dev->tx_queue_len, GFP_KERNEL)) {
+		err = -ENOMEM;
+		goto out;
+	}
+
 	tfile->queue_index = tun->numqueues;
 	tfile->socket.sk->sk_shutdown &= ~RCV_SHUTDOWN;
 	rcu_assign_pointer(tfile->tun, tun);
@@ -891,8 +907,8 @@ static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
 
 	nf_reset(skb);
 
-	/* Enqueue packet */
-	skb_queue_tail(&tfile->socket.sk->sk_receive_queue, skb);
+	if (skb_array_produce(&tfile->tx_array, skb))
+		goto drop;
 
 	/* Notify and wake up reader process */
 	if (tfile->flags & TUN_FASYNC)
@@ -1107,7 +1123,7 @@ static unsigned int tun_chr_poll(struct file *file, poll_table *wait)
 
 	poll_wait(file, sk_sleep(sk), wait);
 
-	if (!skb_queue_empty(&sk->sk_receive_queue))
+	if (!skb_array_empty(&tfile->tx_array))
 		mask |= POLLIN | POLLRDNORM;
 
 	if (sock_writeable(sk) ||
@@ -1426,22 +1442,61 @@ done:
 	return total;
 }
 
+static struct sk_buff *tun_ring_recv(struct tun_file *tfile, int noblock,
+				     int *err)
+{
+	DECLARE_WAITQUEUE(wait, current);
+	struct sk_buff *skb = NULL;
+
+	skb = skb_array_consume(&tfile->tx_array);
+	if (skb)
+		goto out;
+	if (noblock) {
+		*err = -EAGAIN;
+		goto out;
+	}
+
+	add_wait_queue(&tfile->wq.wait, &wait);
+	current->state = TASK_INTERRUPTIBLE;
+
+	while (1) {
+		skb = skb_array_consume(&tfile->tx_array);
+		if (skb)
+			break;
+		if (signal_pending(current)) {
+			*err = -ERESTARTSYS;
+			break;
+		}
+		if (tfile->socket.sk->sk_shutdown & RCV_SHUTDOWN) {
+			*err = -EFAULT;
+			break;
+		}
+
+		schedule();
+	}
+
+	current->state = TASK_RUNNING;
+	remove_wait_queue(&tfile->wq.wait, &wait);
+
+out:
+	return skb;
+}
+
 static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
 			   struct iov_iter *to,
 			   int noblock)
 {
 	struct sk_buff *skb;
 	ssize_t ret;
-	int peeked, err, off = 0;
+	int err;
 
 	tun_debug(KERN_INFO, tun, "tun_do_read\n");
 
 	if (!iov_iter_count(to))
 		return 0;
 
-	/* Read frames from queue */
-	skb = __skb_recv_datagram(tfile->socket.sk, noblock ? MSG_DONTWAIT : 0,
-				  &peeked, &off, &err);
+	/* Read frames from ring */
+	skb = tun_ring_recv(tfile, noblock, &err);
 	if (!skb)
 		return err;
 
@@ -1574,8 +1629,25 @@ out:
 	return ret;
 }
 
+static int tun_peek_len(struct socket *sock)
+{
+	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
+	struct tun_struct *tun;
+	int ret = 0;
+
+	tun = __tun_get(tfile);
+	if (!tun)
+		return 0;
+
+	ret = skb_array_peek_len(&tfile->tx_array);
+	tun_put(tun);
+
+	return ret;
+}
+
 /* Ops structure to mimic raw sockets with tun */
 static const struct proto_ops tun_socket_ops = {
+	.peek_len = tun_peek_len,
 	.sendmsg = tun_sendmsg,
 	.recvmsg = tun_recvmsg,
 };
@@ -2397,6 +2469,53 @@ static const struct ethtool_ops tun_ethtool_ops = {
 	.get_ts_info	= ethtool_op_get_ts_info,
 };
 
+static int tun_queue_resize(struct tun_struct *tun)
+{
+	struct net_device *dev = tun->dev;
+	struct tun_file *tfile;
+	struct skb_array **arrays;
+	int n = tun->numqueues + tun->numdisabled;
+	int ret, i;
+
+	arrays = kmalloc(sizeof *arrays * n, GFP_KERNEL);
+	if (!arrays)
+		return -ENOMEM;
+
+	for (i = 0; i < tun->numqueues; i++) {
+		tfile = rtnl_dereference(tun->tfiles[i]);
+		arrays[i] = &tfile->tx_array;
+	}
+	list_for_each_entry(tfile, &tun->disabled, next)
+		arrays[i++] = &tfile->tx_array;
+
+	ret = skb_array_resize_multiple(arrays, n,
+					dev->tx_queue_len, GFP_KERNEL);
+
+	kfree(arrays);
+	return ret;
+}
+
+static int tun_device_event(struct notifier_block *unused,
+			    unsigned long event, void *ptr)
+{
+	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
+	struct tun_struct *tun = netdev_priv(dev);
+
+	switch (event) {
+	case NETDEV_CHANGE_TX_QUEUE_LEN:
+		if (tun_queue_resize(tun))
+			return NOTIFY_BAD;
+		break;
+	default:
+		break;
+	}
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block tun_notifier_block __read_mostly = {
+	.notifier_call	= tun_device_event,
+};
 
 static int __init tun_init(void)
 {
@@ -2416,6 +2535,8 @@ static int __init tun_init(void)
 		pr_err("Can't register misc device %d\n", TUN_MINOR);
 		goto err_misc;
 	}
+
+	register_netdevice_notifier(&tun_notifier_block);
 	return  0;
 err_misc:
 	rtnl_link_unregister(&tun_link_ops);
@@ -2427,6 +2548,7 @@ static void tun_cleanup(void)
 {
 	misc_deregister(&tun_miscdev);
 	rtnl_link_unregister(&tun_link_ops);
+	unregister_netdevice_notifier(&tun_notifier_block);
 }
 
 /* Get an underlying socket object from tun file.  Returns error unless file is
diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index 1d3e45f..e032ca3 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -481,10 +481,14 @@ out:
 
 static int peek_head_len(struct sock *sk)
 {
+	struct socket *sock = sk->sk_socket;
 	struct sk_buff *head;
 	int len = 0;
 	unsigned long flags;
 
+	if (sock->ops->peek_len)
+		return sock->ops->peek_len(sock);
+
 	spin_lock_irqsave(&sk->sk_receive_queue.lock, flags);
 	head = skb_peek(&sk->sk_receive_queue);
 	if (likely(head)) {
@@ -497,6 +501,16 @@ static int peek_head_len(struct sock *sk)
 	return len;
 }
 
+static int sk_has_rx_data(struct sock *sk)
+{
+	struct socket *sock = sk->sk_socket;
+
+	if (sock->ops->peek_len)
+		return sock->ops->peek_len(sock);
+
+	return skb_queue_empty(&sk->sk_receive_queue);
+}
+
 static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk)
 {
 	struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
@@ -513,7 +527,7 @@ static int vhost_net_rx_peek_head_len(struct vhost_net *net, struct sock *sk)
 		endtime = busy_clock() + vq->busyloop_timeout;
 
 		while (vhost_can_busy_poll(&net->dev, endtime) &&
-		       skb_queue_empty(&sk->sk_receive_queue) &&
+		       !sk_has_rx_data(sk) &&
 		       vhost_vq_avail_empty(&net->dev, vq))
 			cpu_relax_lowlatency();
 
diff --git a/include/linux/net.h b/include/linux/net.h
index 25aa03b..b9f0ff4 100644
--- a/include/linux/net.h
+++ b/include/linux/net.h
@@ -185,6 +185,7 @@ struct proto_ops {
 	ssize_t 	(*splice_read)(struct socket *sock,  loff_t *ppos,
 				       struct pipe_inode_info *pipe, size_t len, unsigned int flags);
 	int		(*set_peek_off)(struct sock *sk, int val);
+	int		(*peek_len)(struct socket *sock);
 };
 
 #define DECLARE_SOCKADDR(type, dst, src)	\
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index e84d9d2..7dc2ec7 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -2237,6 +2237,7 @@ struct netdev_lag_lower_state_info {
 #define NETDEV_PRECHANGEUPPER	0x001A
 #define NETDEV_CHANGELOWERSTATE	0x001B
 #define NETDEV_UDP_TUNNEL_PUSH_INFO	0x001C
+#define NETDEV_CHANGE_TX_QUEUE_LEN	0x001E
 
 int register_netdevice_notifier(struct notifier_block *nb);
 int unregister_netdevice_notifier(struct notifier_block *nb);
diff --git a/include/linux/ptr_ring.h b/include/linux/ptr_ring.h
index 562a65e..2052011 100644
--- a/include/linux/ptr_ring.h
+++ b/include/linux/ptr_ring.h
@@ -102,7 +102,7 @@ static inline bool ptr_ring_full_bh(struct ptr_ring *r)
  */
 static inline int __ptr_ring_produce(struct ptr_ring *r, void *ptr)
 {
-	if (r->queue[r->producer])
+	if (unlikely(!r->size) || r->queue[r->producer])
 		return -ENOSPC;
 
 	r->queue[r->producer++] = ptr;
@@ -164,7 +164,9 @@ static inline int ptr_ring_produce_bh(struct ptr_ring *r, void *ptr)
  */
 static inline void *__ptr_ring_peek(struct ptr_ring *r)
 {
-	return r->queue[r->consumer];
+	if (likely(r->size))
+		return r->queue[r->consumer];
+	return NULL;
 }
 
 /* Note: callers invoking this in a loop must use a compiler barrier,
@@ -347,20 +349,14 @@ static inline int ptr_ring_init(struct ptr_ring *r, int size, gfp_t gfp)
 	return 0;
 }
 
-static inline int ptr_ring_resize(struct ptr_ring *r, int size, gfp_t gfp,
-				  void (*destroy)(void *))
+static inline void **__ptr_ring_swap_queue(struct ptr_ring *r, void **queue,
+					   int size, gfp_t gfp,
+					   void (*destroy)(void *))
 {
-	unsigned long flags;
 	int producer = 0;
-	void **queue = __ptr_ring_init_queue_alloc(size, gfp);
 	void **old;
 	void *ptr;
 
-	if (!queue)
-		return -ENOMEM;
-
-	spin_lock_irqsave(&(r)->producer_lock, flags);
-
 	while ((ptr = ptr_ring_consume(r)))
 		if (producer < size)
 			queue[producer++] = ptr;
@@ -373,6 +369,23 @@ static inline int ptr_ring_resize(struct ptr_ring *r, int size, gfp_t gfp,
 	old = r->queue;
 	r->queue = queue;
 
+	return old;
+}
+
+static inline int ptr_ring_resize(struct ptr_ring *r, int size, gfp_t gfp,
+				  void (*destroy)(void *))
+{
+	unsigned long flags;
+	void **queue = __ptr_ring_init_queue_alloc(size, gfp);
+	void **old;
+
+	if (!queue)
+		return -ENOMEM;
+
+	spin_lock_irqsave(&(r)->producer_lock, flags);
+
+	old = __ptr_ring_swap_queue(r, queue, size, gfp, destroy);
+
 	spin_unlock_irqrestore(&(r)->producer_lock, flags);
 
 	kfree(old);
@@ -380,6 +393,48 @@ static inline int ptr_ring_resize(struct ptr_ring *r, int size, gfp_t gfp,
 	return 0;
 }
 
+static inline int ptr_ring_resize_multiple(struct ptr_ring **rings, int nrings,
+					   int size,
+					   gfp_t gfp, void (*destroy)(void *))
+{
+	unsigned long flags;
+	void ***queues;
+	int i;
+
+	queues = kmalloc(nrings * sizeof *queues, gfp);
+	if (!queues)
+		goto noqueues;
+
+	for (i = 0; i < nrings; ++i) {
+		queues[i] = __ptr_ring_init_queue_alloc(size, gfp);
+		if (!queues[i])
+			goto nomem;
+	}
+
+	for (i = 0; i < nrings; ++i) {
+		spin_lock_irqsave(&(rings[i])->producer_lock, flags);
+		queues[i] = __ptr_ring_swap_queue(rings[i], queues[i],
+						  size, gfp, destroy);
+		spin_unlock_irqrestore(&(rings[i])->producer_lock, flags);
+	}
+
+	for (i = 0; i < nrings; ++i)
+		kfree(queues[i]);
+
+	kfree(queues);
+
+	return 0;
+
+nomem:
+	while (--i >= 0)
+		kfree(queues[i]);
+
+	kfree(queues);
+
+noqueues:
+	return -ENOMEM;
+}
+
 static inline void ptr_ring_cleanup(struct ptr_ring *r, void (*destroy)(void *))
 {
 	void *ptr;
diff --git a/include/linux/skb_array.h b/include/linux/skb_array.h
index 678bfbf..f4dfade 100644
--- a/include/linux/skb_array.h
+++ b/include/linux/skb_array.h
@@ -151,16 +151,25 @@ static inline int skb_array_init(struct skb_array *a, int size, gfp_t gfp)
 	return ptr_ring_init(&a->ring, size, gfp);
 }
 
-void __skb_array_destroy_skb(void *ptr)
+static void __skb_array_destroy_skb(void *ptr)
 {
 	kfree_skb(ptr);
 }
 
-int skb_array_resize(struct skb_array *a, int size, gfp_t gfp)
+static inline int skb_array_resize(struct skb_array *a, int size, gfp_t gfp)
 {
 	return ptr_ring_resize(&a->ring, size, gfp, __skb_array_destroy_skb);
 }
 
+static inline int skb_array_resize_multiple(struct skb_array **rings,
+					    int nrings, int size, gfp_t gfp)
+{
+	BUILD_BUG_ON(offsetof(struct skb_array, ring));
+	return ptr_ring_resize_multiple((struct ptr_ring **)rings,
+					nrings, size, gfp,
+					__skb_array_destroy_skb);
+}
+
 static inline void skb_array_cleanup(struct skb_array *a)
 {
 	ptr_ring_cleanup(&a->ring, __skb_array_destroy_skb);
diff --git a/net/core/net-sysfs.c b/net/core/net-sysfs.c
index 7a0b616..6e4f347 100644
--- a/net/core/net-sysfs.c
+++ b/net/core/net-sysfs.c
@@ -322,7 +322,20 @@ NETDEVICE_SHOW_RW(flags, fmt_hex);
 
 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
 {
-	dev->tx_queue_len = new_len;
+	int res, orig_len = dev->tx_queue_len;
+
+	if (new_len != orig_len) {
+		dev->tx_queue_len = new_len;
+		res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
+		res = notifier_to_errno(res);
+		if (res) {
+			netdev_err(dev,
+				   "refused to change device tx_queue_len\n");
+			dev->tx_queue_len = orig_len;
+			return -EFAULT;
+		}
+	}
+
 	return 0;
 }
 
diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
index cfed7bc..a9e3805 100644
--- a/net/core/rtnetlink.c
+++ b/net/core/rtnetlink.c
@@ -1927,11 +1927,19 @@ static int do_setlink(const struct sk_buff *skb,
 
 	if (tb[IFLA_TXQLEN]) {
 		unsigned long value = nla_get_u32(tb[IFLA_TXQLEN]);
-
-		if (dev->tx_queue_len ^ value)
+		unsigned long orig_len = dev->tx_queue_len;
+
+		if (dev->tx_queue_len ^ value) {
+			dev->tx_queue_len = value;
+			err = call_netdevice_notifiers(
+			      NETDEV_CHANGE_TX_QUEUE_LEN, dev);
+			err = notifier_to_errno(err);
+			if (err) {
+				dev->tx_queue_len = orig_len;
+				goto errout;
+			}
 			status |= DO_SETLINK_NOTIFY;
-
-		dev->tx_queue_len = value;
+		}
 	}
 
 	if (tb[IFLA_OPERSTATE])
diff --git a/tools/virtio/ringtest/ptr_ring.c b/tools/virtio/ringtest/ptr_ring.c
index 74abd74..68e4f9f 100644
--- a/tools/virtio/ringtest/ptr_ring.c
+++ b/tools/virtio/ringtest/ptr_ring.c
@@ -17,6 +17,11 @@
 typedef pthread_spinlock_t  spinlock_t;
 
 typedef int gfp_t;
+static void *kmalloc(unsigned size, gfp_t gfp)
+{
+	return memalign(64, size);
+}
+
 static void *kzalloc(unsigned size, gfp_t gfp)
 {
 	void *p = memalign(64, size);