[RFC PATCH v10 00/16] Provide a zero-copy method on KVM virtio-net.

[RFC PATCH v10 00/16] Provide a zero-copy method on KVM virtio-net.

[xiaohui.xin]

We provide an zero-copy method which driver side may get external
buffers to DMA. Here external means driver don't use kernel space
to allocate skb buffers. Currently the external buffer can be from
guest virtio-net driver.

The idea is simple, just to pin the guest VM user space and then
let host NIC driver has the chance to directly DMA to it. 
The patches are based on vhost-net backend driver. We add a device
which provides proto_ops as sendmsg/recvmsg to vhost-net to
send/recv directly to/from the NIC driver. KVM guest who use the
vhost-net backend may bind any ethX interface in the host side to
get copyless data transfer thru guest virtio-net frontend.

patch 01-10:  	net core and kernel changes.
patch 11-13:  	new device as interface to mantpulate external buffers.
patch 14: 	for vhost-net.
patch 15:	An example on modifying NIC driver to using napi_gro_frags().
patch 16:	An example how to get guest buffers based on driver
		who using napi_gro_frags().

The guest virtio-net driver submits multiple requests thru vhost-net
backend driver to the kernel. And the requests are queued and then
completed after corresponding actions in h/w are done.

For read, user space buffers are dispensed to NIC driver for rx when
a page constructor API is invoked. Means NICs can allocate user buffers
from a page constructor. We add a hook in netif_receive_skb() function
to intercept the incoming packets, and notify the zero-copy device.

For write, the zero-copy deivce may allocates a new host skb and puts
payload on the skb_shinfo(skb)->frags, and copied the header to skb->data.
The request remains pending until the skb is transmitted by h/w.

We provide multiple submits and asynchronous notifiicaton to 
vhost-net too.

Our goal is to improve the bandwidth and reduce the CPU usage.
Exact performance data will be provided later.

What we have not done yet:
	Performance tuning

what we have done in v1:
	polish the RCU usage
	deal with write logging in asynchroush mode in vhost
	add notifier block for mp device
	rename page_ctor to mp_port in netdevice.h to make it looks generic
	add mp_dev_change_flags() for mp device to change NIC state
	add CONIFG_VHOST_MPASSTHRU to limit the usage when module is not load
	a small fix for missing dev_put when fail
	using dynamic minor instead of static minor number
	a __KERNEL__ protect to mp_get_sock()

what we have done in v2:
	
	remove most of the RCU usage, since the ctor pointer is only
	changed by BIND/UNBIND ioctl, and during that time, NIC will be
	stopped to get good cleanup(all outstanding requests are finished),
	so the ctor pointer cannot be raced into wrong situation.

	Remove the struct vhost_notifier with struct kiocb.
	Let vhost-net backend to alloc/free the kiocb and transfer them
	via sendmsg/recvmsg.

	use get_user_pages_fast() and set_page_dirty_lock() when read.

	Add some comments for netdev_mp_port_prep() and handle_mpassthru().

what we have done in v3:
	the async write logging is rewritten 
	a drafted synchronous write function for qemu live migration
	a limit for locked pages from get_user_pages_fast() to prevent Dos
	by using RLIMIT_MEMLOCK
	

what we have done in v4:
	add iocb completion callback from vhost-net to queue iocb in mp device
	replace vq->receiver by mp_sock_data_ready()
	remove stuff in mp device which access structures from vhost-net
	modify skb_reserve() to ignore host NIC driver reserved space
	rebase to the latest vhost tree
	split large patches into small pieces, especially for net core part.
	

what we have done in v5:
	address Arnd Bergmann's comments
		-remove IFF_MPASSTHRU_EXCL flag in mp device
		-Add CONFIG_COMPAT macro
		-remove mp_release ops
	move dev_is_mpassthru() as inline func
	fix a bug in memory relinquish
	Apply to current git (2.6.34-rc6) tree.

what we have done in v6:
	move create_iocb() out of page_dtor which may happen in interrupt context
	-This remove the potential issues which lock called in interrupt context
	make the cache used by mp, vhost as static, and created/destoryed during
	modules init/exit functions.
	-This makes multiple mp guest created at the same time.

what we have done in v7:
	some cleanup prepared to suppprt PS mode

what we have done in v8:
	discarding the modifications to point skb->data to guest buffer directly.
	Add code to modify driver to support napi_gro_frags() with Herbert's comments.
	To support PS mode.
	Add mergeable buffer support in mp device.
	Add GSO/GRO support in mp deice.
	Address comments from Eric Dumazet about cache line and rcu usage.

what we have done in v9:
	v8 patch is based on a fix in dev_gro_receive().
	But Herbert did not agree with the fix we have sent out.
	And he suggest another fix. v9 is modified to base on that fix.
	

what we have done in v10:
	Fix a partial csum error.
	Cleanup some unused fields with struct page_info{} in mp device.
	Modify kmem_cache_zalloc() to kmem_cache_alloc() based on Michael S. Thirkin.

Performance:
	We have seen the performance data request from mailling-list.
	And we are now looking into this.

[RFC PATCH v10 00/16] Provide a zero-copy method on KVM virtio-net.

[xiaohui.xin]

We provide an zero-copy method which driver side may get external
buffers to DMA. Here external means driver don't use kernel space
to allocate skb buffers. Currently the external buffer can be from
guest virtio-net driver.

The idea is simple, just to pin the guest VM user space and then
let host NIC driver has the chance to directly DMA to it. 
The patches are based on vhost-net backend driver. We add a device
which provides proto_ops as sendmsg/recvmsg to vhost-net to
send/recv directly to/from the NIC driver. KVM guest who use the
vhost-net backend may bind any ethX interface in the host side to
get copyless data transfer thru guest virtio-net frontend.

patch 01-10:  	net core and kernel changes.
patch 11-13:  	new device as interface to mantpulate external buffers.
patch 14: 	for vhost-net.
patch 15:	An example on modifying NIC driver to using napi_gro_frags().
patch 16:	An example how to get guest buffers based on driver
		who using napi_gro_frags().

The guest virtio-net driver submits multiple requests thru vhost-net
backend driver to the kernel. And the requests are queued and then
completed after corresponding actions in h/w are done.

For read, user space buffers are dispensed to NIC driver for rx when
a page constructor API is invoked. Means NICs can allocate user buffers
from a page constructor. We add a hook in netif_receive_skb() function
to intercept the incoming packets, and notify the zero-copy device.

For write, the zero-copy deivce may allocates a new host skb and puts
payload on the skb_shinfo(skb)->frags, and copied the header to skb->data.
The request remains pending until the skb is transmitted by h/w.

We provide multiple submits and asynchronous notifiicaton to 
vhost-net too.

Our goal is to improve the bandwidth and reduce the CPU usage.
Exact performance data will be provided later.

What we have not done yet:
	Performance tuning

what we have done in v1:
	polish the RCU usage
	deal with write logging in asynchroush mode in vhost
	add notifier block for mp device
	rename page_ctor to mp_port in netdevice.h to make it looks generic
	add mp_dev_change_flags() for mp device to change NIC state
	add CONIFG_VHOST_MPASSTHRU to limit the usage when module is not load
	a small fix for missing dev_put when fail
	using dynamic minor instead of static minor number
	a __KERNEL__ protect to mp_get_sock()

what we have done in v2:
	
	remove most of the RCU usage, since the ctor pointer is only
	changed by BIND/UNBIND ioctl, and during that time, NIC will be
	stopped to get good cleanup(all outstanding requests are finished),
	so the ctor pointer cannot be raced into wrong situation.

	Remove the struct vhost_notifier with struct kiocb.
	Let vhost-net backend to alloc/free the kiocb and transfer them
	via sendmsg/recvmsg.

	use get_user_pages_fast() and set_page_dirty_lock() when read.

	Add some comments for netdev_mp_port_prep() and handle_mpassthru().

what we have done in v3:
	the async write logging is rewritten 
	a drafted synchronous write function for qemu live migration
	a limit for locked pages from get_user_pages_fast() to prevent Dos
	by using RLIMIT_MEMLOCK
	

what we have done in v4:
	add iocb completion callback from vhost-net to queue iocb in mp device
	replace vq->receiver by mp_sock_data_ready()
	remove stuff in mp device which access structures from vhost-net
	modify skb_reserve() to ignore host NIC driver reserved space
	rebase to the latest vhost tree
	split large patches into small pieces, especially for net core part.
	

what we have done in v5:
	address Arnd Bergmann's comments
		-remove IFF_MPASSTHRU_EXCL flag in mp device
		-Add CONFIG_COMPAT macro
		-remove mp_release ops
	move dev_is_mpassthru() as inline func
	fix a bug in memory relinquish
	Apply to current git (2.6.34-rc6) tree.

what we have done in v6:
	move create_iocb() out of page_dtor which may happen in interrupt context
	-This remove the potential issues which lock called in interrupt context
	make the cache used by mp, vhost as static, and created/destoryed during
	modules init/exit functions.
	-This makes multiple mp guest created at the same time.

what we have done in v7:
	some cleanup prepared to suppprt PS mode

what we have done in v8:
	discarding the modifications to point skb->data to guest buffer directly.
	Add code to modify driver to support napi_gro_frags() with Herbert's comments.
	To support PS mode.
	Add mergeable buffer support in mp device.
	Add GSO/GRO support in mp deice.
	Address comments from Eric Dumazet about cache line and rcu usage.

what we have done in v9:
	v8 patch is based on a fix in dev_gro_receive().
	But Herbert did not agree with the fix we have sent out.
	And he suggest another fix. v9 is modified to base on that fix.
	

what we have done in v10:
	Fix a partial csum error.
	Cleanup some unused fields with struct page_info{} in mp device.
	Modify kmem_cache_zalloc() to kmem_cache_alloc() based on Michael S. Thirkin.

Performance:
	We have seen the performance data request from mailling-list.
	And we are now looking into this.

[RFC PATCH v10 01/16] Add a new structure for skb buffer from external.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/skbuff.h |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 124f90c..74af06c 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -203,6 +203,15 @@ struct skb_shared_info {
 	void *		destructor_arg;
 };
 
+/* The structure is for a skb which pages may point to
+ * an external buffer, which is not allocated from kernel space.
+ * It also contains a destructor for itself.
+ */
+struct skb_ext_page {
+	struct		page *page;
+	void		(*dtor)(struct skb_ext_page *);
+};
+
 /* We divide dataref into two halves.  The higher 16 bits hold references
  * to the payload part of skb->data.  The lower 16 bits hold references to
  * the entire skb->data.  A clone of a headerless skb holds the length of
-- 
1.5.4.4

[RFC PATCH v10 01/16] Add a new structure for skb buffer from external.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/skbuff.h |    9 +++++++++
 1 files changed, 9 insertions(+), 0 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 124f90c..74af06c 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -203,6 +203,15 @@ struct skb_shared_info {
 	void *		destructor_arg;
 };
 
+/* The structure is for a skb which pages may point to
+ * an external buffer, which is not allocated from kernel space.
+ * It also contains a destructor for itself.
+ */
+struct skb_ext_page {
+	struct		page *page;
+	void		(*dtor)(struct skb_ext_page *);
+};
+
 /* We divide dataref into two halves.  The higher 16 bits hold references
  * to the payload part of skb->data.  The lower 16 bits hold references to
  * the entire skb->data.  A clone of a headerless skb holds the length of
-- 
1.5.4.4

[RFC PATCH v10 02/16] Add a new struct for device to manipulate external buffer.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/netdevice.h |   22 +++++++++++++++++++++-
 1 files changed, 21 insertions(+), 1 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index fa8b476..ba582e1 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -530,6 +530,25 @@ struct netdev_queue {
 	unsigned long		tx_dropped;
 } ____cacheline_aligned_in_smp;
 
+/* Add a structure in structure net_device, the new field is
+ * named as mp_port. It's for mediate passthru (zero-copy).
+ * It contains the capability for the net device driver,
+ * a socket, and an external buffer creator, external means
+ * skb buffer belongs to the device may not be allocated from
+ * kernel space.
+ */
+struct mpassthru_port	{
+	int		hdr_len;
+	int		data_len;
+	int		npages;
+	unsigned	flags;
+	struct socket	*sock;
+	int		vnet_hlen;
+	struct skb_ext_page *(*ctor)(struct mpassthru_port *,
+				struct sk_buff *, int);
+	struct skb_ext_page *(*hash)(struct net_device *,
+				struct page *);
+};
 
 /*
  * This structure defines the management hooks for network devices.
@@ -952,7 +971,8 @@ struct net_device {
 	struct macvlan_port	*macvlan_port;
 	/* GARP */
 	struct garp_port	*garp_port;
-
+	/* mpassthru */
+	struct mpassthru_port	*mp_port;
 	/* class/net/name entry */
 	struct device		dev;
 	/* space for optional device, statistics, and wireless sysfs groups */
-- 
1.5.4.4

[RFC PATCH v10 02/16] Add a new struct for device to manipulate external buffer.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/netdevice.h |   22 +++++++++++++++++++++-
 1 files changed, 21 insertions(+), 1 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index fa8b476..ba582e1 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -530,6 +530,25 @@ struct netdev_queue {
 	unsigned long		tx_dropped;
 } ____cacheline_aligned_in_smp;
 
+/* Add a structure in structure net_device, the new field is
+ * named as mp_port. It's for mediate passthru (zero-copy).
+ * It contains the capability for the net device driver,
+ * a socket, and an external buffer creator, external means
+ * skb buffer belongs to the device may not be allocated from
+ * kernel space.
+ */
+struct mpassthru_port	{
+	int		hdr_len;
+	int		data_len;
+	int		npages;
+	unsigned	flags;
+	struct socket	*sock;
+	int		vnet_hlen;
+	struct skb_ext_page *(*ctor)(struct mpassthru_port *,
+				struct sk_buff *, int);
+	struct skb_ext_page *(*hash)(struct net_device *,
+				struct page *);
+};
 
 /*
  * This structure defines the management hooks for network devices.
@@ -952,7 +971,8 @@ struct net_device {
 	struct macvlan_port	*macvlan_port;
 	/* GARP */
 	struct garp_port	*garp_port;
-
+	/* mpassthru */
+	struct mpassthru_port	*mp_port;
 	/* class/net/name entry */
 	struct device		dev;
 	/* space for optional device, statistics, and wireless sysfs groups */
-- 
1.5.4.4

[RFC PATCH v10 03/16] Add a ndo_mp_port_prep pointer to net_device_ops.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

If the driver want to allocate external buffers,
then it can export it's capability, as the skb
buffer header length, the page length can be DMA, etc.
The external buffers owner may utilize this.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/netdevice.h |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index ba582e1..aba0308 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -710,6 +710,10 @@ struct net_device_ops {
 	int			(*ndo_fcoe_get_wwn)(struct net_device *dev,
 						    u64 *wwn, int type);
 #endif
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+	int			(*ndo_mp_port_prep)(struct net_device *dev,
+						struct mpassthru_port *port);
+#endif
 };
 
 /*
-- 
1.5.4.4

[RFC PATCH v10 03/16] Add a ndo_mp_port_prep pointer to net_device_ops.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

If the driver want to allocate external buffers,
then it can export it's capability, as the skb
buffer header length, the page length can be DMA, etc.
The external buffers owner may utilize this.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/netdevice.h |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index ba582e1..aba0308 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -710,6 +710,10 @@ struct net_device_ops {
 	int			(*ndo_fcoe_get_wwn)(struct net_device *dev,
 						    u64 *wwn, int type);
 #endif
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+	int			(*ndo_mp_port_prep)(struct net_device *dev,
+						struct mpassthru_port *port);
+#endif
 };
 
 /*
-- 
1.5.4.4

[RFC PATCH v10 04/16] Add a function make external buffer owner to query capability.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

The external buffer owner can use the functions to get
the capability of the underlying NIC driver.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhaonew@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>

---
 include/linux/netdevice.h |    2 +
 net/core/dev.c            |   49 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 51 insertions(+), 0 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index aba0308..5f192de 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1599,6 +1599,8 @@ extern gro_result_t	napi_frags_finish(struct napi_struct *napi,
 					  gro_result_t ret);
 extern struct sk_buff *	napi_frags_skb(struct napi_struct *napi);
 extern gro_result_t	napi_gro_frags(struct napi_struct *napi);
+extern int netdev_mp_port_prep(struct net_device *dev,
+				struct mpassthru_port *port);
 
 static inline void napi_free_frags(struct napi_struct *napi)
 {
diff --git a/net/core/dev.c b/net/core/dev.c
index 264137f..636f11b 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2468,6 +2468,55 @@ void netif_nit_deliver(struct sk_buff *skb)
 	rcu_read_unlock();
 }
 
+/* To support meidate passthru(zero-copy) with NIC driver,
+ * we'd better query NIC driver for the capability it can
+ * provide, especially for packet split mode, now we only
+ * query for the header size, and the payload a descriptor
+ * may carry. If a driver does not use the API to export,
+ * then we may try to use a default value, currently,
+ * we use the default value from an IGB driver. Now,
+ * it's only called by mpassthru device.
+ */
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+int netdev_mp_port_prep(struct net_device *dev,
+		struct mpassthru_port *port)
+{
+	int rc;
+	int npages, data_len;
+	const struct net_device_ops *ops = dev->netdev_ops;
+
+	if (ops->ndo_mp_port_prep) {
+		rc = ops->ndo_mp_port_prep(dev, port);
+		if (rc)
+			return rc;
+	} else {
+		/* If the NIC driver did not report this,
+		 * then we try to use default value.
+		 */
+		port->hdr_len = 128;
+		port->data_len = 2048;
+		port->npages = 1;
+	}
+
+	if (port->hdr_len <= 0)
+		goto err;
+
+	npages = port->npages;
+	data_len = port->data_len;
+	if (npages <= 0 || npages > MAX_SKB_FRAGS ||
+			(data_len < PAGE_SIZE * (npages - 1) ||
+			 data_len > PAGE_SIZE * npages))
+		goto err;
+
+	return 0;
+err:
+	dev_warn(&dev->dev, "invalid page constructor parameters\n");
+
+	return -EINVAL;
+}
+EXPORT_SYMBOL(netdev_mp_port_prep);
+#endif
+
 /**
  *	netif_receive_skb - process receive buffer from network
  *	@skb: buffer to process
-- 
1.5.4.4

[RFC PATCH v10 04/16] Add a function make external buffer owner to query capability.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

The external buffer owner can use the functions to get
the capability of the underlying NIC driver.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhaonew@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>

---
 include/linux/netdevice.h |    2 +
 net/core/dev.c            |   49 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 51 insertions(+), 0 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index aba0308..5f192de 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1599,6 +1599,8 @@ extern gro_result_t	napi_frags_finish(struct napi_struct *napi,
 					  gro_result_t ret);
 extern struct sk_buff *	napi_frags_skb(struct napi_struct *napi);
 extern gro_result_t	napi_gro_frags(struct napi_struct *napi);
+extern int netdev_mp_port_prep(struct net_device *dev,
+				struct mpassthru_port *port);
 
 static inline void napi_free_frags(struct napi_struct *napi)
 {
diff --git a/net/core/dev.c b/net/core/dev.c
index 264137f..636f11b 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2468,6 +2468,55 @@ void netif_nit_deliver(struct sk_buff *skb)
 	rcu_read_unlock();
 }
 
+/* To support meidate passthru(zero-copy) with NIC driver,
+ * we'd better query NIC driver for the capability it can
+ * provide, especially for packet split mode, now we only
+ * query for the header size, and the payload a descriptor
+ * may carry. If a driver does not use the API to export,
+ * then we may try to use a default value, currently,
+ * we use the default value from an IGB driver. Now,
+ * it's only called by mpassthru device.
+ */
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+int netdev_mp_port_prep(struct net_device *dev,
+		struct mpassthru_port *port)
+{
+	int rc;
+	int npages, data_len;
+	const struct net_device_ops *ops = dev->netdev_ops;
+
+	if (ops->ndo_mp_port_prep) {
+		rc = ops->ndo_mp_port_prep(dev, port);
+		if (rc)
+			return rc;
+	} else {
+		/* If the NIC driver did not report this,
+		 * then we try to use default value.
+		 */
+		port->hdr_len = 128;
+		port->data_len = 2048;
+		port->npages = 1;
+	}
+
+	if (port->hdr_len <= 0)
+		goto err;
+
+	npages = port->npages;
+	data_len = port->data_len;
+	if (npages <= 0 || npages > MAX_SKB_FRAGS ||
+			(data_len < PAGE_SIZE * (npages - 1) ||
+			 data_len > PAGE_SIZE * npages))
+		goto err;
+
+	return 0;
+err:
+	dev_warn(&dev->dev, "invalid page constructor parameters\n");
+
+	return -EINVAL;
+}
+EXPORT_SYMBOL(netdev_mp_port_prep);
+#endif
+
 /**
  *	netif_receive_skb - process receive buffer from network
  *	@skb: buffer to process
-- 
1.5.4.4

[RFC PATCH v10 05/16] Add a function to indicate if device use external buffer.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/netdevice.h |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 5f192de..23d6ec0 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1602,6 +1602,11 @@ extern gro_result_t	napi_gro_frags(struct napi_struct *napi);
 extern int netdev_mp_port_prep(struct net_device *dev,
 				struct mpassthru_port *port);
 
+static inline bool dev_is_mpassthru(struct net_device *dev)
+{
+	return dev && dev->mp_port;
+}
+
 static inline void napi_free_frags(struct napi_struct *napi)
 {
 	kfree_skb(napi->skb);
-- 
1.5.4.4

[RFC PATCH v10 05/16] Add a function to indicate if device use external buffer.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/netdevice.h |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 5f192de..23d6ec0 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1602,6 +1602,11 @@ extern gro_result_t	napi_gro_frags(struct napi_struct *napi);
 extern int netdev_mp_port_prep(struct net_device *dev,
 				struct mpassthru_port *port);
 
+static inline bool dev_is_mpassthru(struct net_device *dev)
+{
+	return dev && dev->mp_port;
+}
+
 static inline void napi_free_frags(struct napi_struct *napi)
 {
 	kfree_skb(napi->skb);
-- 
1.5.4.4

[RFC PATCH v10 06/16] Use callback to deal with skb_release_data() specially.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

    If buffer is external, then use the callback to destruct
    buffers.

    Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
    Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
    Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/skbuff.h |    3 ++-
 net/core/skbuff.c      |    8 ++++++++
 2 files changed, 10 insertions(+), 1 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 74af06c..ab29675 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -197,10 +197,11 @@ struct skb_shared_info {
 	union skb_shared_tx tx_flags;
 	struct sk_buff	*frag_list;
 	struct skb_shared_hwtstamps hwtstamps;
-	skb_frag_t	frags[MAX_SKB_FRAGS];
 	/* Intermediate layers must ensure that destructor_arg
 	 * remains valid until skb destructor */
 	void *		destructor_arg;
+
+	skb_frag_t	frags[MAX_SKB_FRAGS];
 };
 
 /* The structure is for a skb which pages may point to
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 93c4e06..117d82b 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -217,6 +217,7 @@ struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 	shinfo->gso_type = 0;
 	shinfo->ip6_frag_id = 0;
 	shinfo->tx_flags.flags = 0;
+	shinfo->destructor_arg = NULL;
 	skb_frag_list_init(skb);
 	memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
 
@@ -350,6 +351,13 @@ static void skb_release_data(struct sk_buff *skb)
 		if (skb_has_frags(skb))
 			skb_drop_fraglist(skb);
 
+		if (skb->dev && dev_is_mpassthru(skb->dev)) {
+			struct skb_ext_page *ext_page =
+				skb_shinfo(skb)->destructor_arg;
+			if (ext_page && ext_page->dtor)
+				ext_page->dtor(ext_page);
+		}
+
 		kfree(skb->head);
 	}
 }
-- 
1.5.4.4

[RFC PATCH v10 06/16] Use callback to deal with skb_release_data() specially.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

    If buffer is external, then use the callback to destruct
    buffers.

    Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
    Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
    Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/skbuff.h |    3 ++-
 net/core/skbuff.c      |    8 ++++++++
 2 files changed, 10 insertions(+), 1 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 74af06c..ab29675 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -197,10 +197,11 @@ struct skb_shared_info {
 	union skb_shared_tx tx_flags;
 	struct sk_buff	*frag_list;
 	struct skb_shared_hwtstamps hwtstamps;
-	skb_frag_t	frags[MAX_SKB_FRAGS];
 	/* Intermediate layers must ensure that destructor_arg
 	 * remains valid until skb destructor */
 	void *		destructor_arg;
+
+	skb_frag_t	frags[MAX_SKB_FRAGS];
 };
 
 /* The structure is for a skb which pages may point to
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 93c4e06..117d82b 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -217,6 +217,7 @@ struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 	shinfo->gso_type = 0;
 	shinfo->ip6_frag_id = 0;
 	shinfo->tx_flags.flags = 0;
+	shinfo->destructor_arg = NULL;
 	skb_frag_list_init(skb);
 	memset(&shinfo->hwtstamps, 0, sizeof(shinfo->hwtstamps));
 
@@ -350,6 +351,13 @@ static void skb_release_data(struct sk_buff *skb)
 		if (skb_has_frags(skb))
 			skb_drop_fraglist(skb);
 
+		if (skb->dev && dev_is_mpassthru(skb->dev)) {
+			struct skb_ext_page *ext_page =
+				skb_shinfo(skb)->destructor_arg;
+			if (ext_page && ext_page->dtor)
+				ext_page->dtor(ext_page);
+		}
+
 		kfree(skb->head);
 	}
 }
-- 
1.5.4.4

[RFC PATCH v10 07/16] Modify netdev_alloc_page() to get external buffer

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

    Currently, it can get external buffers from mp device.

    Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
    Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
    Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 net/core/skbuff.c |   27 +++++++++++++++++++++++++++
 1 files changed, 27 insertions(+), 0 deletions(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 117d82b..1a61e2b 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -269,11 +269,38 @@ struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
 }
 EXPORT_SYMBOL(__netdev_alloc_skb);
 
+struct page *netdev_alloc_ext_pages(struct net_device *dev, int npages)
+{
+	struct mpassthru_port *port;
+	struct skb_ext_page *ext_page = NULL;
+
+	port = dev->mp_port;
+	if (!port)
+		goto out;
+	ext_page = port->ctor(port, NULL, npages);
+	if (ext_page)
+		return ext_page->page;
+out:
+	return NULL;
+
+}
+EXPORT_SYMBOL(netdev_alloc_ext_pages);
+
+struct page *netdev_alloc_ext_page(struct net_device *dev)
+{
+	return netdev_alloc_ext_pages(dev, 1);
+
+}
+EXPORT_SYMBOL(netdev_alloc_ext_page);
+
 struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
 {
 	int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
 	struct page *page;
 
+	if (dev_is_mpassthru(dev))
+		return netdev_alloc_ext_page(dev);
+
 	page = alloc_pages_node(node, gfp_mask, 0);
 	return page;
 }
-- 
1.5.4.4

[RFC PATCH v10 07/16] Modify netdev_alloc_page() to get external buffer

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

    Currently, it can get external buffers from mp device.

    Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
    Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
    Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 net/core/skbuff.c |   27 +++++++++++++++++++++++++++
 1 files changed, 27 insertions(+), 0 deletions(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 117d82b..1a61e2b 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -269,11 +269,38 @@ struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
 }
 EXPORT_SYMBOL(__netdev_alloc_skb);
 
+struct page *netdev_alloc_ext_pages(struct net_device *dev, int npages)
+{
+	struct mpassthru_port *port;
+	struct skb_ext_page *ext_page = NULL;
+
+	port = dev->mp_port;
+	if (!port)
+		goto out;
+	ext_page = port->ctor(port, NULL, npages);
+	if (ext_page)
+		return ext_page->page;
+out:
+	return NULL;
+
+}
+EXPORT_SYMBOL(netdev_alloc_ext_pages);
+
+struct page *netdev_alloc_ext_page(struct net_device *dev)
+{
+	return netdev_alloc_ext_pages(dev, 1);
+
+}
+EXPORT_SYMBOL(netdev_alloc_ext_page);
+
 struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
 {
 	int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
 	struct page *page;
 
+	if (dev_is_mpassthru(dev))
+		return netdev_alloc_ext_page(dev);
+
 	page = alloc_pages_node(node, gfp_mask, 0);
 	return page;
 }
-- 
1.5.4.4

[RFC PATCH v10 08/16] Modify netdev_free_page() to release external buffer

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Currently, it can get external buffers from mp device.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/skbuff.h |    4 +++-
 net/core/skbuff.c      |   24 ++++++++++++++++++++++++
 2 files changed, 27 insertions(+), 1 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index ab29675..3d7f70e 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -1512,9 +1512,11 @@ static inline struct page *netdev_alloc_page(struct net_device *dev)
 	return __netdev_alloc_page(dev, GFP_ATOMIC);
 }
 
+extern void __netdev_free_page(struct net_device *dev, struct page *page);
+
 static inline void netdev_free_page(struct net_device *dev, struct page *page)
 {
-	__free_page(page);
+	__netdev_free_page(dev, page);
 }
 
 /**
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 1a61e2b..bbf4707 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -306,6 +306,30 @@ struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
 }
 EXPORT_SYMBOL(__netdev_alloc_page);
 
+void netdev_free_ext_page(struct net_device *dev, struct page *page)
+{
+	struct skb_ext_page *ext_page = NULL;
+	if (dev_is_mpassthru(dev) && dev->mp_port->hash) {
+		ext_page = dev->mp_port->hash(dev, page);
+		if (ext_page)
+			ext_page->dtor(ext_page);
+		else
+			__free_page(page);
+	}
+}
+EXPORT_SYMBOL(netdev_free_ext_page);
+
+void __netdev_free_page(struct net_device *dev, struct page *page)
+{
+	if (dev_is_mpassthru(dev)) {
+		netdev_free_ext_page(dev, page);
+		return;
+	}
+
+	__free_page(page);
+}
+EXPORT_SYMBOL(__netdev_free_page);
+
 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 		int size)
 {
-- 
1.5.4.4

[RFC PATCH v10 08/16] Modify netdev_free_page() to release external buffer

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Currently, it can get external buffers from mp device.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/skbuff.h |    4 +++-
 net/core/skbuff.c      |   24 ++++++++++++++++++++++++
 2 files changed, 27 insertions(+), 1 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index ab29675..3d7f70e 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -1512,9 +1512,11 @@ static inline struct page *netdev_alloc_page(struct net_device *dev)
 	return __netdev_alloc_page(dev, GFP_ATOMIC);
 }
 
+extern void __netdev_free_page(struct net_device *dev, struct page *page);
+
 static inline void netdev_free_page(struct net_device *dev, struct page *page)
 {
-	__free_page(page);
+	__netdev_free_page(dev, page);
 }
 
 /**
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 1a61e2b..bbf4707 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -306,6 +306,30 @@ struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
 }
 EXPORT_SYMBOL(__netdev_alloc_page);
 
+void netdev_free_ext_page(struct net_device *dev, struct page *page)
+{
+	struct skb_ext_page *ext_page = NULL;
+	if (dev_is_mpassthru(dev) && dev->mp_port->hash) {
+		ext_page = dev->mp_port->hash(dev, page);
+		if (ext_page)
+			ext_page->dtor(ext_page);
+		else
+			__free_page(page);
+	}
+}
+EXPORT_SYMBOL(netdev_free_ext_page);
+
+void __netdev_free_page(struct net_device *dev, struct page *page)
+{
+	if (dev_is_mpassthru(dev)) {
+		netdev_free_ext_page(dev, page);
+		return;
+	}
+
+	__free_page(page);
+}
+EXPORT_SYMBOL(__netdev_free_page);
+
 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 		int size)
 {
-- 
1.5.4.4

[RFC PATCH v10 09/16] Don't do skb recycle, if device use external buffer.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

    Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
    Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
    Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 net/core/skbuff.c |    6 ++++++
 1 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index bbf4707..9b156bb 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -565,6 +565,12 @@ int skb_recycle_check(struct sk_buff *skb, int skb_size)
 	if (skb_shared(skb) || skb_cloned(skb))
 		return 0;
 
+	/* if the device wants to do mediate passthru, the skb may
+	 * get external buffer, so don't recycle
+	 */
+	if (dev_is_mpassthru(skb->dev))
+		return 0;
+
 	skb_release_head_state(skb);
 	shinfo = skb_shinfo(skb);
 	atomic_set(&shinfo->dataref, 1);
-- 
1.5.4.4

[RFC PATCH v10 09/16] Don't do skb recycle, if device use external buffer.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

    Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
    Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
    Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 net/core/skbuff.c |    6 ++++++
 1 files changed, 6 insertions(+), 0 deletions(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index bbf4707..9b156bb 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -565,6 +565,12 @@ int skb_recycle_check(struct sk_buff *skb, int skb_size)
 	if (skb_shared(skb) || skb_cloned(skb))
 		return 0;
 
+	/* if the device wants to do mediate passthru, the skb may
+	 * get external buffer, so don't recycle
+	 */
+	if (dev_is_mpassthru(skb->dev))
+		return 0;
+
 	skb_release_head_state(skb);
 	shinfo = skb_shinfo(skb);
 	atomic_set(&shinfo->dataref, 1);
-- 
1.5.4.4

[RFC PATCH v10 10/16] Add a hook to intercept external buffers from NIC driver.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

The hook is called in netif_receive_skb().
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 net/core/dev.c |   35 +++++++++++++++++++++++++++++++++++
 1 files changed, 35 insertions(+), 0 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 636f11b..4b379b1 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2517,6 +2517,37 @@ err:
 EXPORT_SYMBOL(netdev_mp_port_prep);
 #endif
 
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+/* Add a hook to intercept mediate passthru(zero-copy) packets,
+ * and insert it to the socket queue owned by mp_port specially.
+ */
+static inline struct sk_buff *handle_mpassthru(struct sk_buff *skb,
+					       struct packet_type **pt_prev,
+					       int *ret,
+					       struct net_device *orig_dev)
+{
+	struct mpassthru_port *mp_port = NULL;
+	struct sock *sk = NULL;
+
+	if (!dev_is_mpassthru(skb->dev))
+		return skb;
+	mp_port = skb->dev->mp_port;
+
+	if (*pt_prev) {
+		*ret = deliver_skb(skb, *pt_prev, orig_dev);
+		*pt_prev = NULL;
+	}
+
+	sk = mp_port->sock->sk;
+	skb_queue_tail(&sk->sk_receive_queue, skb);
+	sk->sk_state_change(sk);
+
+	return NULL;
+}
+#else
+#define handle_mpassthru(skb, pt_prev, ret, orig_dev)     (skb)
+#endif
+
 /**
  *	netif_receive_skb - process receive buffer from network
  *	@skb: buffer to process
@@ -2598,6 +2629,10 @@ int netif_receive_skb(struct sk_buff *skb)
 ncls:
 #endif
 
+	/* To intercept mediate passthru(zero-copy) packets here */
+	skb = handle_mpassthru(skb, &pt_prev, &ret, orig_dev);
+	if (!skb)
+		goto out;
 	skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
 	if (!skb)
 		goto out;
-- 
1.5.4.4

[RFC PATCH v10 10/16] Add a hook to intercept external buffers from NIC driver.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

The hook is called in netif_receive_skb().
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 net/core/dev.c |   35 +++++++++++++++++++++++++++++++++++
 1 files changed, 35 insertions(+), 0 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 636f11b..4b379b1 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2517,6 +2517,37 @@ err:
 EXPORT_SYMBOL(netdev_mp_port_prep);
 #endif
 
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+/* Add a hook to intercept mediate passthru(zero-copy) packets,
+ * and insert it to the socket queue owned by mp_port specially.
+ */
+static inline struct sk_buff *handle_mpassthru(struct sk_buff *skb,
+					       struct packet_type **pt_prev,
+					       int *ret,
+					       struct net_device *orig_dev)
+{
+	struct mpassthru_port *mp_port = NULL;
+	struct sock *sk = NULL;
+
+	if (!dev_is_mpassthru(skb->dev))
+		return skb;
+	mp_port = skb->dev->mp_port;
+
+	if (*pt_prev) {
+		*ret = deliver_skb(skb, *pt_prev, orig_dev);
+		*pt_prev = NULL;
+	}
+
+	sk = mp_port->sock->sk;
+	skb_queue_tail(&sk->sk_receive_queue, skb);
+	sk->sk_state_change(sk);
+
+	return NULL;
+}
+#else
+#define handle_mpassthru(skb, pt_prev, ret, orig_dev)     (skb)
+#endif
+
 /**
  *	netif_receive_skb - process receive buffer from network
  *	@skb: buffer to process
@@ -2598,6 +2629,10 @@ int netif_receive_skb(struct sk_buff *skb)
 ncls:
 #endif
 
+	/* To intercept mediate passthru(zero-copy) packets here */
+	skb = handle_mpassthru(skb, &pt_prev, &ret, orig_dev);
+	if (!skb)
+		goto out;
 	skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
 	if (!skb)
 		goto out;
-- 
1.5.4.4

[RFC PATCH v10 11/16] Add header file for mp device.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/mpassthru.h |   25 +++++++++++++++++++++++++
 1 files changed, 25 insertions(+), 0 deletions(-)
 create mode 100644 include/linux/mpassthru.h

diff --git a/include/linux/mpassthru.h b/include/linux/mpassthru.h
new file mode 100644
index 0000000..ba8f320
--- /dev/null
+++ b/include/linux/mpassthru.h
@@ -0,0 +1,25 @@
+#ifndef __MPASSTHRU_H
+#define __MPASSTHRU_H
+
+#include <linux/types.h>
+#include <linux/if_ether.h>
+
+/* ioctl defines */
+#define MPASSTHRU_BINDDEV      _IOW('M', 213, int)
+#define MPASSTHRU_UNBINDDEV    _IO('M', 214)
+
+#ifdef __KERNEL__
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+struct socket *mp_get_socket(struct file *);
+#else
+#include <linux/err.h>
+#include <linux/errno.h>
+struct file;
+struct socket;
+static inline struct socket *mp_get_socket(struct file *f)
+{
+	return ERR_PTR(-EINVAL);
+}
+#endif /* CONFIG_MEDIATE_PASSTHRU */
+#endif /* __KERNEL__ */
+#endif /* __MPASSTHRU_H */
-- 
1.5.4.4

[RFC PATCH v10 11/16] Add header file for mp device.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 include/linux/mpassthru.h |   25 +++++++++++++++++++++++++
 1 files changed, 25 insertions(+), 0 deletions(-)
 create mode 100644 include/linux/mpassthru.h

diff --git a/include/linux/mpassthru.h b/include/linux/mpassthru.h
new file mode 100644
index 0000000..ba8f320
--- /dev/null
+++ b/include/linux/mpassthru.h
@@ -0,0 +1,25 @@
+#ifndef __MPASSTHRU_H
+#define __MPASSTHRU_H
+
+#include <linux/types.h>
+#include <linux/if_ether.h>
+
+/* ioctl defines */
+#define MPASSTHRU_BINDDEV      _IOW('M', 213, int)
+#define MPASSTHRU_UNBINDDEV    _IO('M', 214)
+
+#ifdef __KERNEL__
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+struct socket *mp_get_socket(struct file *);
+#else
+#include <linux/err.h>
+#include <linux/errno.h>
+struct file;
+struct socket;
+static inline struct socket *mp_get_socket(struct file *f)
+{
+	return ERR_PTR(-EINVAL);
+}
+#endif /* CONFIG_MEDIATE_PASSTHRU */
+#endif /* __KERNEL__ */
+#endif /* __MPASSTHRU_H */
-- 
1.5.4.4

[RFC PATCH v10 12/16] Add a kconfig entry and make entry for mp device.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 drivers/vhost/Kconfig  |   10 ++++++++++
 drivers/vhost/Makefile |    2 ++
 2 files changed, 12 insertions(+), 0 deletions(-)

diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig
index e4e2fd1..a6b8cbf 100644
--- a/drivers/vhost/Kconfig
+++ b/drivers/vhost/Kconfig
@@ -9,3 +9,13 @@ config VHOST_NET
 	  To compile this driver as a module, choose M here: the module will
 	  be called vhost_net.
 
+config MEDIATE_PASSTHRU
+	tristate "mediate passthru network driver (EXPERIMENTAL)"
+	depends on VHOST_NET
+	---help---
+	  zerocopy network I/O support, we call it as mediate passthru to
+	  be distiguish with hardare passthru.
+
+	  To compile this driver as a module, choose M here: the module will
+	  be called mpassthru.
+
diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile
index 72dd020..c18b9fc 100644
--- a/drivers/vhost/Makefile
+++ b/drivers/vhost/Makefile
@@ -1,2 +1,4 @@
 obj-$(CONFIG_VHOST_NET) += vhost_net.o
 vhost_net-y := vhost.o net.o
+
+obj-$(CONFIG_MEDIATE_PASSTHRU) += mpassthru.o
-- 
1.5.4.4

[RFC PATCH v10 12/16] Add a kconfig entry and make entry for mp device.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 drivers/vhost/Kconfig  |   10 ++++++++++
 drivers/vhost/Makefile |    2 ++
 2 files changed, 12 insertions(+), 0 deletions(-)

diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig
index e4e2fd1..a6b8cbf 100644
--- a/drivers/vhost/Kconfig
+++ b/drivers/vhost/Kconfig
@@ -9,3 +9,13 @@ config VHOST_NET
 	  To compile this driver as a module, choose M here: the module will
 	  be called vhost_net.
 
+config MEDIATE_PASSTHRU
+	tristate "mediate passthru network driver (EXPERIMENTAL)"
+	depends on VHOST_NET
+	---help---
+	  zerocopy network I/O support, we call it as mediate passthru to
+	  be distiguish with hardare passthru.
+
+	  To compile this driver as a module, choose M here: the module will
+	  be called mpassthru.
+
diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile
index 72dd020..c18b9fc 100644
--- a/drivers/vhost/Makefile
+++ b/drivers/vhost/Makefile
@@ -1,2 +1,4 @@
 obj-$(CONFIG_VHOST_NET) += vhost_net.o
 vhost_net-y := vhost.o net.o
+
+obj-$(CONFIG_MEDIATE_PASSTHRU) += mpassthru.o
-- 
1.5.4.4

[RFC PATCH v10 13/16] Add mp(mediate passthru) device.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

The patch add mp(mediate passthru) device, which now
based on vhost-net backend driver and provides proto_ops
to send/receive guest buffers data from/to guest vitio-net
driver.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 drivers/vhost/mpassthru.c | 1407 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 1407 insertions(+), 0 deletions(-)
 create mode 100644 drivers/vhost/mpassthru.c

diff --git a/drivers/vhost/mpassthru.c b/drivers/vhost/mpassthru.c
new file mode 100644
index 0000000..d86d94c
--- /dev/null
+++ b/drivers/vhost/mpassthru.c
@@ -0,0 +1,1407 @@
+/*
+ *  MPASSTHRU - Mediate passthrough device.
+ *  Copyright (C) 2009 ZhaoYu, XinXiaohui, Dike, Jeffery G
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ *  GNU General Public License for more details.
+ *
+ */
+
+#define DRV_NAME        "mpassthru"
+#define DRV_DESCRIPTION "Mediate passthru device driver"
+#define DRV_COPYRIGHT   "(C) 2009 ZhaoYu, XinXiaohui, Dike, Jeffery G"
+
+#include <linux/compat.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/major.h>
+#include <linux/slab.h>
+#include <linux/smp_lock.h>
+#include <linux/poll.h>
+#include <linux/fcntl.h>
+#include <linux/init.h>
+#include <linux/aio.h>
+
+#include <linux/skbuff.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/miscdevice.h>
+#include <linux/ethtool.h>
+#include <linux/rtnetlink.h>
+#include <linux/if.h>
+#include <linux/if_arp.h>
+#include <linux/if_ether.h>
+#include <linux/crc32.h>
+#include <linux/nsproxy.h>
+#include <linux/uaccess.h>
+#include <linux/virtio_net.h>
+#include <linux/mpassthru.h>
+#include <net/net_namespace.h>
+#include <net/netns/generic.h>
+#include <net/rtnetlink.h>
+#include <net/sock.h>
+
+#include <asm/system.h>
+
+/* Uncomment to enable debugging */
+/* #define MPASSTHRU_DEBUG 1 */
+
+#ifdef MPASSTHRU_DEBUG
+static int debug;
+
+#define DBG  if (mp->debug) printk
+#define DBG1 if (debug == 2) printk
+#else
+#define DBG(a...)
+#define DBG1(a...)
+#endif
+
+#define COPY_THRESHOLD (L1_CACHE_BYTES * 4)
+#define COPY_HDR_LEN   (L1_CACHE_BYTES < 64 ? 64 : L1_CACHE_BYTES)
+
+struct frag {
+	u16     offset;
+	u16     size;
+};
+
+#define	HASH_BUCKETS	(8192*2)
+
+struct page_info {
+	struct list_head        list;
+	struct page_info	*next;
+	struct page_info	*prev;
+	struct page             *pages[MAX_SKB_FRAGS];
+	struct sk_buff          *skb;
+	struct page_ctor        *ctor;
+
+	/* The pointer relayed to skb, to indicate
+	 * it's a external allocated skb or kernel
+	 */
+	struct skb_ext_page    ext_page;
+
+#define INFO_READ                      0
+#define INFO_WRITE                     1
+	unsigned                flags;
+	unsigned                pnum;
+
+	/* The fields after that is for backend
+	 * driver, now for vhost-net.
+	 */
+
+	struct kiocb            *iocb;
+	unsigned int            desc_pos;
+	struct iovec            hdr[2];
+	struct iovec            iov[MAX_SKB_FRAGS];
+};
+
+static struct kmem_cache *ext_page_info_cache;
+
+struct page_ctor {
+	struct list_head        readq;
+	int			wq_len;
+	int			rq_len;
+	spinlock_t		read_lock;
+	/* record the locked pages */
+	int			lock_pages;
+	struct rlimit		o_rlim;
+	struct net_device	*dev;
+	struct mpassthru_port	port;
+	struct page_info	**hash_table;
+};
+
+struct mp_struct {
+	struct mp_file		*mfile;
+	struct net_device       *dev;
+	struct page_ctor	*ctor;
+	struct socket           socket;
+
+#ifdef MPASSTHRU_DEBUG
+	int debug;
+#endif
+};
+
+struct mp_file {
+	atomic_t count;
+	struct mp_struct *mp;
+	struct net *net;
+};
+
+struct mp_sock {
+	struct sock		sk;
+	struct mp_struct	*mp;
+};
+
+static int mp_dev_change_flags(struct net_device *dev, unsigned flags)
+{
+	int ret = 0;
+
+	rtnl_lock();
+	ret = dev_change_flags(dev, flags);
+	rtnl_unlock();
+
+	if (ret < 0)
+		printk(KERN_ERR "failed to change dev state of %s", dev->name);
+
+	return ret;
+}
+
+/* The main function to allocate external buffers */
+static struct skb_ext_page *page_ctor(struct mpassthru_port *port,
+		struct sk_buff *skb, int npages)
+{
+	int i;
+	unsigned long flags;
+	struct page_ctor *ctor;
+	struct page_info *info = NULL;
+
+	ctor = container_of(port, struct page_ctor, port);
+
+	spin_lock_irqsave(&ctor->read_lock, flags);
+	if (!list_empty(&ctor->readq)) {
+		info = list_first_entry(&ctor->readq, struct page_info, list);
+		list_del(&info->list);
+		ctor->rq_len--;
+	}
+	spin_unlock_irqrestore(&ctor->read_lock, flags);
+	if (!info)
+		return NULL;
+
+	for (i = 0; i < info->pnum; i++)
+		get_page(info->pages[i]);
+	info->skb = skb;
+	return &info->ext_page;
+}
+
+static struct page_info *mp_hash_lookup(struct page_ctor *ctor,
+					struct page *page);
+static struct page_info *mp_hash_delete(struct page_ctor *ctor,
+					struct page_info *info);
+
+static struct skb_ext_page *mp_lookup(struct net_device *dev,
+				      struct page *page)
+{
+	struct mp_struct *mp =
+		container_of(dev->mp_port->sock->sk, struct mp_sock, sk)->mp;
+	struct page_ctor *ctor = mp->ctor;
+	struct page_info *info;
+
+	info = mp_hash_lookup(ctor, page);
+	if (!info)
+		return NULL;
+	return &info->ext_page;
+}
+
+static int page_ctor_attach(struct mp_struct *mp)
+{
+	int rc;
+	struct page_ctor *ctor;
+	struct net_device *dev = mp->dev;
+
+	/* locked by mp_mutex */
+	if (mp->ctor)
+		return -EBUSY;
+
+	ctor = kzalloc(sizeof(*ctor), GFP_KERNEL);
+	if (!ctor)
+		return -ENOMEM;
+	rc = netdev_mp_port_prep(dev, &ctor->port);
+	if (rc)
+		goto fail;
+
+	INIT_LIST_HEAD(&ctor->readq);
+	spin_lock_init(&ctor->read_lock);
+	ctor->hash_table = kzalloc(sizeof(struct page_info *) * HASH_BUCKETS,
+			GFP_KERNEL);
+	if (!ctor->hash_table)
+		goto fail_hash;
+
+	ctor->rq_len = 0;
+	ctor->wq_len = 0;
+
+	dev_hold(dev);
+	ctor->dev = dev;
+	ctor->port.ctor = page_ctor;
+	ctor->port.sock = &mp->socket;
+	ctor->port.hash = mp_lookup;
+	ctor->lock_pages = 0;
+
+	/* locked by mp_mutex */
+	dev->mp_port = &ctor->port;
+	mp->ctor = ctor;
+
+	return 0;
+
+fail_hash:
+	kfree(ctor->hash_table);
+
+fail:
+	kfree(ctor);
+	dev_put(dev);
+
+	return rc;
+}
+
+struct page_info *info_dequeue(struct page_ctor *ctor)
+{
+	unsigned long flags;
+	struct page_info *info = NULL;
+	spin_lock_irqsave(&ctor->read_lock, flags);
+	if (!list_empty(&ctor->readq)) {
+		info = list_first_entry(&ctor->readq,
+				struct page_info, list);
+		list_del(&info->list);
+		ctor->rq_len--;
+	}
+	spin_unlock_irqrestore(&ctor->read_lock, flags);
+	return info;
+}
+
+static int set_memlock_rlimit(struct page_ctor *ctor, int resource,
+			      unsigned long cur, unsigned long max)
+{
+	struct rlimit new_rlim, *old_rlim;
+	int retval;
+
+	if (resource != RLIMIT_MEMLOCK)
+		return -EINVAL;
+	new_rlim.rlim_cur = cur;
+	new_rlim.rlim_max = max;
+
+	old_rlim = current->signal->rlim + resource;
+
+	/* remember the old rlimit value when backend enabled */
+	ctor->o_rlim.rlim_cur = old_rlim->rlim_cur;
+	ctor->o_rlim.rlim_max = old_rlim->rlim_max;
+
+	if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
+			!capable(CAP_SYS_RESOURCE))
+		return -EPERM;
+
+	retval = security_task_setrlimit(resource, &new_rlim);
+	if (retval)
+		return retval;
+
+	task_lock(current->group_leader);
+	*old_rlim = new_rlim;
+	task_unlock(current->group_leader);
+	return 0;
+}
+
+static void relinquish_resource(struct page_ctor *ctor)
+{
+	if (!(ctor->dev->flags & IFF_UP) &&
+			!(ctor->wq_len + ctor->rq_len))
+		printk(KERN_INFO "relinquish_resource\n");
+}
+
+static void mp_ki_dtor(struct kiocb *iocb)
+{
+	struct page_info *info = (struct page_info *)(iocb->private);
+	int i;
+
+	if (info->flags == INFO_READ) {
+		for (i = 0; i < info->pnum; i++) {
+			if (info->pages[i]) {
+				set_page_dirty_lock(info->pages[i]);
+				put_page(info->pages[i]);
+			}
+		}
+		mp_hash_delete(info->ctor, info);
+		if (info->skb) {
+			info->skb->destructor = NULL;
+			kfree_skb(info->skb);
+		}
+		info->ctor->rq_len--;
+	} else
+		info->ctor->wq_len--;
+	/* Decrement the number of locked pages */
+	info->ctor->lock_pages -= info->pnum;
+	kmem_cache_free(ext_page_info_cache, info);
+	relinquish_resource(info->ctor);
+
+	return;
+}
+
+static struct kiocb *create_iocb(struct page_info *info, int size)
+{
+	struct kiocb *iocb = NULL;
+
+	iocb = info->iocb;
+	if (!iocb)
+		return iocb;
+	iocb->ki_flags = 0;
+	iocb->ki_users = 1;
+	iocb->ki_key = 0;
+	iocb->ki_ctx = NULL;
+	iocb->ki_cancel = NULL;
+	iocb->ki_retry = NULL;
+	iocb->ki_eventfd = NULL;
+	iocb->ki_pos = info->desc_pos;
+	iocb->ki_nbytes = size;
+	iocb->ki_dtor(iocb);
+	iocb->private = (void *)info;
+	iocb->ki_dtor = mp_ki_dtor;
+
+	return iocb;
+}
+
+static int page_ctor_detach(struct mp_struct *mp)
+{
+	struct page_ctor *ctor;
+	struct page_info *info;
+	int i;
+
+	/* locked by mp_mutex */
+	ctor = mp->ctor;
+	if (!ctor)
+		return -ENODEV;
+
+	while ((info = info_dequeue(ctor))) {
+		for (i = 0; i < info->pnum; i++)
+			if (info->pages[i])
+				put_page(info->pages[i]);
+		create_iocb(info, 0);
+		ctor->rq_len--;
+		kmem_cache_free(ext_page_info_cache, info);
+	}
+
+	relinquish_resource(ctor);
+
+	set_memlock_rlimit(ctor, RLIMIT_MEMLOCK,
+			   ctor->o_rlim.rlim_cur,
+			   ctor->o_rlim.rlim_max);
+
+	/* locked by mp_mutex */
+	ctor->dev->mp_port = NULL;
+	dev_put(ctor->dev);
+
+	mp->ctor = NULL;
+	kfree(ctor->hash_table);
+	kfree(ctor);
+	return 0;
+}
+
+static void __mp_detach(struct mp_struct *mp)
+{
+	mp->mfile = NULL;
+
+	mp_dev_change_flags(mp->dev, mp->dev->flags & ~IFF_UP);
+	page_ctor_detach(mp);
+	mp_dev_change_flags(mp->dev, mp->dev->flags | IFF_UP);
+
+	/* Drop the extra count on the net device */
+	dev_put(mp->dev);
+}
+
+static DEFINE_MUTEX(mp_mutex);
+
+static void mp_detach(struct mp_struct *mp)
+{
+	mutex_lock(&mp_mutex);
+	__mp_detach(mp);
+	mutex_unlock(&mp_mutex);
+}
+
+static struct mp_struct *mp_get(struct mp_file *mfile)
+{
+	struct mp_struct *mp = NULL;
+	if (atomic_inc_not_zero(&mfile->count))
+		mp = mfile->mp;
+
+	return mp;
+}
+
+static void mp_put(struct mp_file *mfile)
+{
+	if (atomic_dec_and_test(&mfile->count))
+		mp_detach(mfile->mp);
+}
+
+static void iocb_tag(struct kiocb *iocb)
+{
+	iocb->ki_flags = 1;
+}
+
+/* The callback to destruct the external buffers or skb */
+static void page_dtor(struct skb_ext_page *ext_page)
+{
+	struct page_info *info;
+	struct page_ctor *ctor;
+	struct sock *sk;
+	struct sk_buff *skb;
+
+	if (!ext_page)
+		return;
+	info = container_of(ext_page, struct page_info, ext_page);
+	if (!info)
+		return;
+	ctor = info->ctor;
+	skb = info->skb;
+
+	if (info->flags == INFO_READ) {
+		create_iocb(info, 0);
+		return;
+	}
+
+	/* For transmit, we should wait for the DMA finish by hardware.
+	 * Queue the notifier to wake up the backend driver
+	 */
+
+	iocb_tag(info->iocb);
+	sk = ctor->port.sock->sk;
+	sk->sk_write_space(sk);
+
+	return;
+}
+
+/* For small exteranl buffers transmit, we don't need to call
+ * get_user_pages().
+ */
+static struct page_info *alloc_small_page_info(struct page_ctor *ctor,
+		struct kiocb *iocb, int total)
+{
+	struct page_info *info =
+		kmem_cache_alloc(ext_page_info_cache, GFP_KERNEL);
+
+	if (!info)
+		return NULL;
+	info->ext_page.dtor = page_dtor;
+	info->ctor = ctor;
+	info->flags = INFO_WRITE;
+	info->iocb = iocb;
+	info->pnum = 0;
+	return info;
+}
+
+typedef u32 key_mp_t;
+static inline key_mp_t mp_hash(struct page *page, int buckets)
+{
+	key_mp_t k;
+
+	k = ((((unsigned long)page << 32UL) >> 32UL) / 0x38) % buckets ;
+	return k;
+}
+
+static void mp_hash_insert(struct page_ctor *ctor,
+		struct page *page, struct page_info *page_info)
+{
+	struct page_info *tmp;
+	key_mp_t key = mp_hash(page, HASH_BUCKETS);
+	if (!ctor->hash_table[key]) {
+		ctor->hash_table[key] = page_info;
+		return;
+	}
+
+	tmp = ctor->hash_table[key];
+	while (tmp->next)
+		tmp = tmp->next;
+
+	tmp->next = page_info;
+	page_info->prev = tmp;
+	return;
+}
+
+static struct page_info *mp_hash_delete(struct page_ctor *ctor,
+					struct page_info *info)
+{
+	key_mp_t key = mp_hash(info->pages[0], HASH_BUCKETS);
+	struct page_info *tmp = NULL;
+	int i;
+
+	tmp = ctor->hash_table[key];
+	while (tmp) {
+		if (tmp == info) {
+			if (!tmp->prev) {
+				ctor->hash_table[key] = tmp->next;
+				if (tmp->next)
+					tmp->next->prev = NULL;
+			} else {
+				tmp->prev->next = tmp->next;
+				if (tmp->next)
+					tmp->next->prev = tmp->prev;
+			}
+			return tmp;
+		}
+		tmp = tmp->next;
+	}
+	return tmp;
+}
+
+static struct page_info *mp_hash_lookup(struct page_ctor *ctor,
+					struct page *page)
+{
+	key_mp_t key = mp_hash(page, HASH_BUCKETS);
+	struct page_info *tmp = NULL;
+
+	int i;
+	tmp = ctor->hash_table[key];
+	while (tmp) {
+		for (i = 0; i < tmp->pnum; i++) {
+			if (tmp->pages[i] == page)
+				return tmp;
+		}
+		tmp = tmp->next;
+	}
+	return tmp;
+}
+
+/* The main function to transform the guest user space address
+ * to host kernel address via get_user_pages(). Thus the hardware
+ * can do DMA directly to the external buffer address.
+ */
+static struct page_info *alloc_page_info(struct page_ctor *ctor,
+		struct kiocb *iocb, struct iovec *iov,
+		int count, struct frag *frags,
+		int npages, int total)
+{
+	int rc;
+	int i, j, n = 0;
+	int len;
+	unsigned long base, lock_limit;
+	struct page_info *info = NULL;
+
+	lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
+	lock_limit >>= PAGE_SHIFT;
+
+	if (ctor->lock_pages + count > lock_limit && npages) {
+		printk(KERN_INFO "exceed the locked memory rlimit.");
+		return NULL;
+	}
+
+	info = kmem_cache_alloc(ext_page_info_cache, GFP_KERNEL);
+	
+	if (!info)
+		return NULL;
+	info->skb = NULL;
+	info->next = info->prev = NULL;
+
+	for (i = j = 0; i < count; i++) {
+		base = (unsigned long)iov[i].iov_base;
+		len = iov[i].iov_len;
+
+		if (!len)
+			continue;
+		n = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
+
+		rc = get_user_pages_fast(base, n, npages ? 1 : 0,
+				&info->pages[j]);
+		if (rc != n)
+			goto failed;
+
+		while (n--) {
+			frags[j].offset = base & ~PAGE_MASK;
+			frags[j].size = min_t(int, len,
+					PAGE_SIZE - frags[j].offset);
+			len -= frags[j].size;
+			base += frags[j].size;
+			j++;
+		}
+	}
+
+#ifdef CONFIG_HIGHMEM
+	if (npages && !(dev->features & NETIF_F_HIGHDMA)) {
+		for (i = 0; i < j; i++) {
+			if (PageHighMem(info->pages[i]))
+				goto failed;
+		}
+	}
+#endif
+
+	info->ext_page.dtor = page_dtor;
+	info->ext_page.page = info->pages[0];
+	info->ctor = ctor;
+	info->pnum = j;
+	info->iocb = iocb;
+	if (!npages)
+		info->flags = INFO_WRITE;
+	else
+		info->flags = INFO_READ;
+
+	if (info->flags == INFO_READ) {
+		if (frags[0].offset == 0 && iocb->ki_iovec[0].iov_len) {
+			frags[0].offset = iocb->ki_iovec[0].iov_len;
+			ctor->port.vnet_hlen = iocb->ki_iovec[0].iov_len;
+		}
+		for (i = 0; i < j; i++)
+			mp_hash_insert(ctor, info->pages[i], info);
+	}
+	/* increment the number of locked pages */
+	ctor->lock_pages += j;
+	return info;
+
+failed:
+	for (i = 0; i < j; i++)
+		put_page(info->pages[i]);
+
+	kmem_cache_free(ext_page_info_cache, info);
+
+	return NULL;
+}
+
+static void mp_sock_destruct(struct sock *sk)
+{
+	struct mp_struct *mp = container_of(sk, struct mp_sock, sk)->mp;
+	kfree(mp);
+}
+
+static void mp_sock_state_change(struct sock *sk)
+{
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN);
+}
+
+static void mp_sock_write_space(struct sock *sk)
+{
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT);
+}
+
+static void mp_sock_data_ready(struct sock *sk, int coming)
+{
+	struct mp_struct *mp = container_of(sk, struct mp_sock, sk)->mp;
+	struct page_ctor *ctor = NULL;
+	struct sk_buff *skb = NULL;
+	struct page_info *info = NULL;
+	int len;
+
+	ctor = mp->ctor;
+	if (!ctor)
+		return;
+
+	while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
+		struct page *page;
+		int off;
+		int size = 0, i = 0;
+		struct skb_shared_info *shinfo = skb_shinfo(skb);
+		struct skb_ext_page *ext_page =
+			(struct skb_ext_page *)(shinfo->destructor_arg);
+		struct virtio_net_hdr_mrg_rxbuf hdr = {
+			.hdr.flags = 0,
+			.hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE
+		};
+
+		if (skb->ip_summed == CHECKSUM_COMPLETE)
+			DBG(KERN_INFO "csum %d\n", skb->ip_summed);
+
+		if (shinfo->frags[0].page == ext_page->page) {
+			info = container_of(ext_page,
+					    struct page_info,
+					    ext_page);
+			if (shinfo->nr_frags)
+				hdr.num_buffers = shinfo->nr_frags;
+			else
+				hdr.num_buffers = shinfo->nr_frags + 1;
+		} else {
+			info = container_of(ext_page,
+					    struct page_info,
+					    ext_page);
+			hdr.num_buffers = shinfo->nr_frags + 1;
+		}
+		skb_push(skb, ETH_HLEN);
+
+		if (skb_is_gso(skb)) {
+			hdr.hdr.hdr_len = skb_headlen(skb);
+			hdr.hdr.gso_size = shinfo->gso_size;
+			if (shinfo->gso_type & SKB_GSO_TCPV4)
+				hdr.hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+			else if (shinfo->gso_type & SKB_GSO_TCPV6)
+				hdr.hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
+			else if (shinfo->gso_type & SKB_GSO_UDP)
+				hdr.hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
+			else
+				BUG();
+			if (shinfo->gso_type & SKB_GSO_TCP_ECN)
+				hdr.hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
+
+		} else
+			hdr.hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
+
+		if (skb->ip_summed == CHECKSUM_PARTIAL) {
+			hdr.hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
+			hdr.hdr.csum_start =
+				skb->csum_start - skb_headroom(skb);
+			hdr.hdr.csum_offset = skb->csum_offset;
+		}
+
+		off = info->hdr[0].iov_len;
+		len = memcpy_toiovec(info->iov, (unsigned char *)&hdr, off);
+		if (len) {
+			DBG(KERN_INFO
+				"Unable to write vnet_hdr at addr %p: %d\n",
+				info->iov_base, len);
+			goto clean;
+		}
+
+		memcpy_toiovec(info->iov, skb->data, skb_headlen(skb));
+
+		info->iocb->ki_left = hdr.num_buffers;
+		if (shinfo->frags[0].page == ext_page->page) {
+			size = shinfo->frags[0].size +
+				shinfo->frags[0].page_offset - off;
+			i = 1;
+		} else {
+			size = skb_headlen(skb);
+			i = 0;
+		}
+		create_iocb(info, off + size);
+		for (i = i; i < shinfo->nr_frags; i++) {
+			page = shinfo->frags[i].page;
+			info = mp_hash_lookup(ctor, shinfo->frags[i].page);
+			create_iocb(info, shinfo->frags[i].size);
+		}
+		info->skb = skb;
+		shinfo->nr_frags = 0;
+		shinfo->destructor_arg = NULL;
+		continue;
+clean:
+		kfree_skb(skb);
+		for (i = 0; i < info->pnum; i++)
+			put_page(info->pages[i]);
+		kmem_cache_free(ext_page_info_cache, info);
+	}
+	return;
+}
+
+static inline struct sk_buff *mp_alloc_skb(struct sock *sk, size_t prepad,
+					   size_t len, size_t linear,
+					   int noblock, int *err)
+{
+	struct sk_buff *skb;
+
+	/* Under a page?  Don't bother with paged skb. */
+	if (prepad + len < PAGE_SIZE || !linear)
+		linear = len;
+
+	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
+			err);
+	if (!skb)
+		return NULL;
+
+	skb_reserve(skb, prepad);
+	skb_put(skb, linear);
+	skb->data_len = len - linear;
+	skb->len += len - linear;
+
+	return skb;
+}
+
+static int mp_skb_from_vnet_hdr(struct sk_buff *skb,
+		struct virtio_net_hdr *vnet_hdr)
+{
+	unsigned short gso_type = 0;
+	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+		switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
+		case VIRTIO_NET_HDR_GSO_TCPV4:
+			gso_type = SKB_GSO_TCPV4;
+			break;
+		case VIRTIO_NET_HDR_GSO_TCPV6:
+			gso_type = SKB_GSO_TCPV6;
+			break;
+		case VIRTIO_NET_HDR_GSO_UDP:
+			gso_type = SKB_GSO_UDP;
+			break;
+		default:
+			return -EINVAL;
+		}
+
+		if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
+			gso_type |= SKB_GSO_TCP_ECN;
+
+		if (vnet_hdr->gso_size == 0)
+			return -EINVAL;
+	}
+
+	if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
+		if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
+					vnet_hdr->csum_offset))
+			return -EINVAL;
+	}
+
+	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+		skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
+		skb_shinfo(skb)->gso_type = gso_type;
+
+		/* Header must be checked, and gso_segs computed. */
+		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
+		skb_shinfo(skb)->gso_segs = 0;
+	}
+	return 0;
+}
+
+static int mp_sendmsg(struct kiocb *iocb, struct socket *sock,
+		struct msghdr *m, size_t total_len)
+{
+	struct mp_struct *mp = container_of(sock->sk, struct mp_sock, sk)->mp;
+	struct virtio_net_hdr vnet_hdr = {0};
+	int hdr_len = 0;
+	struct page_ctor *ctor;
+	struct iovec *iov = m->msg_iov;
+	struct page_info *info = NULL;
+	struct frag frags[MAX_SKB_FRAGS];
+	struct sk_buff *skb;
+	int count = m->msg_iovlen;
+	int total = 0, header, n, i, len, rc;
+	unsigned long base;
+
+	ctor = mp->ctor;
+	if (!ctor)
+		return -ENODEV;
+
+	total = iov_length(iov, count);
+
+	if (total < ETH_HLEN)
+		return -EINVAL;
+
+	if (total <= COPY_THRESHOLD)
+		goto copy;
+
+	n = 0;
+	for (i = 0; i < count; i++) {
+		base = (unsigned long)iov[i].iov_base;
+		len = iov[i].iov_len;
+		if (!len)
+			continue;
+		n += ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
+		if (n > MAX_SKB_FRAGS)
+			return -EINVAL;
+	}
+
+copy:
+	hdr_len = sizeof(vnet_hdr);
+	if ((total - iocb->ki_iovec[0].iov_len) < 0)
+		return -EINVAL;
+
+	rc = memcpy_fromiovecend((void *)&vnet_hdr, iocb->ki_iovec, 0, hdr_len);
+	if (rc < 0)
+		return -EINVAL;
+
+	if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
+			vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
+			vnet_hdr.hdr_len)
+		vnet_hdr.hdr_len = vnet_hdr.csum_start +
+			vnet_hdr.csum_offset + 2;
+
+	if (vnet_hdr.hdr_len > total)
+		return -EINVAL;
+
+	header = total > COPY_THRESHOLD ? COPY_HDR_LEN : total;
+
+	skb = mp_alloc_skb(sock->sk, NET_IP_ALIGN, header,
+			   iocb->ki_iovec[0].iov_len, 1, &rc);
+
+	if (!skb)
+		goto drop;
+
+	skb_set_network_header(skb, ETH_HLEN);
+	memcpy_fromiovec(skb->data, iov, header);
+
+	skb_reset_mac_header(skb);
+	skb->protocol = eth_hdr(skb)->h_proto;
+
+	rc = mp_skb_from_vnet_hdr(skb, &vnet_hdr);
+	if (rc)
+		goto drop;
+
+	if (header == total) {
+		rc = total;
+		info = alloc_small_page_info(ctor, iocb, total);
+	} else {
+		info = alloc_page_info(ctor, iocb, iov, count, frags, 0, total);
+		if (info)
+			for (i = 0; i < info->pnum; i++) {
+				skb_add_rx_frag(skb, i, info->pages[i],
+						frags[i].offset, frags[i].size);
+				info->pages[i] = NULL;
+			}
+	}
+	if (info != NULL) {
+		info->desc_pos = iocb->ki_pos;
+		info->skb = skb;
+		skb_shinfo(skb)->destructor_arg = &info->ext_page;
+		skb->dev = mp->dev;
+		ctor->wq_len++;
+		create_iocb(info, total);
+		dev_queue_xmit(skb);
+		if (!ctor->rq_len)
+			sock->sk->sk_state_change(sock->sk);
+		return 0;
+	}
+drop:
+	kfree_skb(skb);
+	if (info) {
+		for (i = 0; i < info->pnum; i++)
+			put_page(info->pages[i]);
+		kmem_cache_free(ext_page_info_cache, info);
+	}
+	mp->dev->stats.tx_dropped++;
+	return -ENOMEM;
+}
+
+static int mp_recvmsg(struct kiocb *iocb, struct socket *sock,
+		struct msghdr *m, size_t total_len,
+		int flags)
+{
+	struct mp_struct *mp = container_of(sock->sk, struct mp_sock, sk)->mp;
+	struct page_ctor *ctor;
+	struct iovec *iov = m->msg_iov;
+	int count = m->msg_iovlen;
+	int npages, payload;
+	struct page_info *info;
+	struct frag frags[MAX_SKB_FRAGS];
+	unsigned long base;
+	int i, len;
+	unsigned long flag;
+
+	if (!(flags & MSG_DONTWAIT))
+		return -EINVAL;
+
+	ctor = mp->ctor;
+	if (!ctor)
+		return -EINVAL;
+
+	/* Error detections in case invalid external buffer */
+	if (count > 2 && iov[1].iov_len < ctor->port.hdr_len &&
+			mp->dev->features & NETIF_F_SG) {
+		return -EINVAL;
+	}
+
+	npages = ctor->port.npages;
+	payload = ctor->port.data_len;
+
+	/* If KVM guest virtio-net FE driver use SG feature */
+	if (count > 2) {
+		for (i = 2; i < count; i++) {
+			base = (unsigned long)iov[i].iov_base & ~PAGE_MASK;
+			len = iov[i].iov_len;
+			if (npages == 1)
+				len = min_t(int, len, PAGE_SIZE - base);
+			else if (base)
+				break;
+			payload -= len;
+			if (payload <= 0)
+				goto proceed;
+			if (npages == 1 || (len & ~PAGE_MASK))
+				break;
+		}
+	}
+
+	if ((((unsigned long)iov[1].iov_base & ~PAGE_MASK)
+				- NET_SKB_PAD - NET_IP_ALIGN) >= 0)
+		goto proceed;
+
+	return -EINVAL;
+
+proceed:
+	/* skip the virtnet head */
+	if (count > 1) {
+		iov++;
+		count--;
+	}
+
+	if (!ctor->lock_pages || !ctor->rq_len) {
+		set_memlock_rlimit(ctor, RLIMIT_MEMLOCK,
+				iocb->ki_user_data * 4096 * 2,
+				iocb->ki_user_data * 4096 * 2);
+	}
+
+	/* Translate address to kernel */
+	info = alloc_page_info(ctor, iocb, iov, count, frags, npages, 0);
+	if (!info)
+		return -ENOMEM;
+	info->hdr[0].iov_base = iocb->ki_iovec[0].iov_base;
+	info->hdr[0].iov_len = iocb->ki_iovec[0].iov_len;
+	iocb->ki_iovec[0].iov_len = 0;
+	iocb->ki_left = 0;
+	info->desc_pos = iocb->ki_pos;
+
+	if (count > 1) {
+		iov--;
+		count++;
+	}
+
+	memcpy(info->iov, iov, sizeof(struct iovec) * count);
+
+	spin_lock_irqsave(&ctor->read_lock, flag);
+	list_add_tail(&info->list, &ctor->readq);
+	spin_unlock_irqrestore(&ctor->read_lock, flag);
+
+	ctor->rq_len++;
+
+	return 0;
+}
+
+/* Ops structure to mimic raw sockets with mp device */
+static const struct proto_ops mp_socket_ops = {
+	.sendmsg = mp_sendmsg,
+	.recvmsg = mp_recvmsg,
+};
+
+static struct proto mp_proto = {
+	.name           = "mp",
+	.owner          = THIS_MODULE,
+	.obj_size       = sizeof(struct mp_sock),
+};
+
+static int mp_chr_open(struct inode *inode, struct file * file)
+{
+	struct mp_file *mfile;
+	cycle_kernel_lock();
+	DBG1(KERN_INFO "mp: mp_chr_open\n");
+
+	mfile = kzalloc(sizeof(*mfile), GFP_KERNEL);
+	if (!mfile)
+		return -ENOMEM;
+	atomic_set(&mfile->count, 0);
+	mfile->mp = NULL;
+	mfile->net = get_net(current->nsproxy->net_ns);
+	file->private_data = mfile;
+	return 0;
+}
+
+static int mp_attach(struct mp_struct *mp, struct file *file)
+{
+	struct mp_file *mfile = file->private_data;
+	int err;
+
+	netif_tx_lock_bh(mp->dev);
+
+	err = -EINVAL;
+
+	if (mfile->mp)
+		goto out;
+
+	err = -EBUSY;
+	if (mp->mfile)
+		goto out;
+
+	err = 0;
+	mfile->mp = mp;
+	mp->mfile = mfile;
+	mp->socket.file = file;
+	dev_hold(mp->dev);
+	sock_hold(mp->socket.sk);
+	atomic_inc(&mfile->count);
+
+out:
+	netif_tx_unlock_bh(mp->dev);
+	return err;
+}
+
+static int do_unbind(struct mp_file *mfile)
+{
+	struct mp_struct *mp = mp_get(mfile);
+
+	if (!mp)
+		return -EINVAL;
+
+	mp_detach(mp);
+	sock_put(mp->socket.sk);
+	mp_put(mfile);
+	return 0;
+}
+
+static long mp_chr_ioctl(struct file *file, unsigned int cmd,
+		unsigned long arg)
+{
+	struct mp_file *mfile = file->private_data;
+	struct mp_struct *mp;
+	struct net_device *dev;
+	void __user* argp = (void __user *)arg;
+	struct ifreq ifr;
+	struct sock *sk;
+	int ret;
+
+	ret = -EINVAL;
+
+	switch (cmd) {
+	case MPASSTHRU_BINDDEV:
+		ret = -EFAULT;
+		if (copy_from_user(&ifr, argp, sizeof ifr))
+			break;
+
+		ifr.ifr_name[IFNAMSIZ-1] = '\0';
+
+		ret = -ENODEV;
+		dev = dev_get_by_name(mfile->net, ifr.ifr_name);
+		if (!dev)
+			break;
+
+		mutex_lock(&mp_mutex);
+
+		ret = -EBUSY;
+
+		/* the device can be only bind once */
+		if (dev_is_mpassthru(dev))
+			goto err_dev_put;
+
+		mp = mfile->mp;
+		if (mp)
+			goto err_dev_put;
+
+		mp = kzalloc(sizeof(*mp), GFP_KERNEL);
+		if (!mp) {
+			ret = -ENOMEM;
+			goto err_dev_put;
+		}
+		mp->dev = dev;
+		ret = -ENOMEM;
+
+		sk = sk_alloc(mfile->net, AF_UNSPEC, GFP_KERNEL, &mp_proto);
+		if (!sk)
+			goto err_free_mp;
+
+		init_waitqueue_head(&mp->socket.wait);
+		mp->socket.ops = &mp_socket_ops;
+		sock_init_data(&mp->socket, sk);
+		sk->sk_sndbuf = INT_MAX;
+		container_of(sk, struct mp_sock, sk)->mp = mp;
+
+		sk->sk_destruct = mp_sock_destruct;
+		sk->sk_data_ready = mp_sock_data_ready;
+		sk->sk_write_space = mp_sock_write_space;
+		sk->sk_state_change = mp_sock_state_change;
+		ret = mp_attach(mp, file);
+		if (ret < 0)
+			goto err_free_sk;
+
+		ret = page_ctor_attach(mp);
+		if (ret < 0)
+			goto err_free_sk;
+		mp_dev_change_flags(mp->dev, mp->dev->flags & (~IFF_UP));
+		mp_dev_change_flags(mp->dev, mp->dev->flags | IFF_UP);
+		sk->sk_state_change(sk);
+out:
+		mutex_unlock(&mp_mutex);
+		break;
+err_free_sk:
+		sk_free(sk);
+err_free_mp:
+		kfree(mp);
+err_dev_put:
+		dev_put(dev);
+		goto out;
+
+	case MPASSTHRU_UNBINDDEV:
+		ret = do_unbind(mfile);
+		break;
+
+	default:
+		break;
+	}
+	return ret;
+}
+
+static unsigned int mp_chr_poll(struct file *file, poll_table * wait)
+{
+	struct mp_file *mfile = file->private_data;
+	struct mp_struct *mp = mp_get(mfile);
+	struct sock *sk;
+	unsigned int mask = 0;
+
+	if (!mp)
+		return POLLERR;
+
+	sk = mp->socket.sk;
+
+	poll_wait(file, &mp->socket.wait, wait);
+
+	if (!skb_queue_empty(&sk->sk_receive_queue))
+		mask |= POLLIN | POLLRDNORM;
+
+	if (sock_writeable(sk) ||
+		(!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
+			 sock_writeable(sk)))
+		mask |= POLLOUT | POLLWRNORM;
+
+	if (mp->dev->reg_state != NETREG_REGISTERED)
+		mask = POLLERR;
+
+	mp_put(mfile);
+	return mask;
+}
+
+static ssize_t mp_chr_aio_write(struct kiocb *iocb, const struct iovec *iov,
+				unsigned long count, loff_t pos)
+{
+	struct file *file = iocb->ki_filp;
+	struct mp_struct *mp = mp_get(file->private_data);
+	struct sock *sk = mp->socket.sk;
+	struct sk_buff *skb;
+	int len, err;
+	ssize_t result = 0;
+
+	if (!mp)
+		return -EBADFD;
+
+	/* currently, async is not supported.
+	 * but we may support real async aio from user application,
+	 * maybe qemu virtio-net backend.
+	 */
+	if (!is_sync_kiocb(iocb))
+		return -EFAULT;
+
+	len = iov_length(iov, count);
+
+	if (unlikely(len) < ETH_HLEN)
+		return -EINVAL;
+
+	skb = sock_alloc_send_skb(sk, len + NET_IP_ALIGN,
+				  file->f_flags & O_NONBLOCK, &err);
+
+	if (!skb)
+		return -EFAULT;
+
+	skb_reserve(skb, NET_IP_ALIGN);
+	skb_put(skb, len);
+
+	if (skb_copy_datagram_from_iovec(skb, 0, iov, 0, len)) {
+		kfree_skb(skb);
+		return -EAGAIN;
+	}
+
+	skb->protocol = eth_type_trans(skb, mp->dev);
+	skb->dev = mp->dev;
+
+	dev_queue_xmit(skb);
+
+	mp_put(file->private_data);
+	return result;
+}
+
+static int mp_chr_close(struct inode *inode, struct file *file)
+{
+	struct mp_file *mfile = file->private_data;
+
+	/*
+	 * Ignore return value since an error only means there was nothing to
+	 * do
+	 */
+	do_unbind(mfile);
+
+	put_net(mfile->net);
+	kfree(mfile);
+
+	return 0;
+}
+
+#ifdef CONFIG_COMPAT
+static long mp_chr_compat_ioctl(struct file *f, unsigned int ioctl,
+				unsigned long arg)
+{
+	return mp_chr_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
+}
+#endif
+
+static const struct file_operations mp_fops = {
+	.owner  = THIS_MODULE,
+	.llseek = no_llseek,
+	.write  = do_sync_write,
+	.aio_write = mp_chr_aio_write,
+	.poll   = mp_chr_poll,
+	.unlocked_ioctl = mp_chr_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl = mp_chr_compat_ioctl,
+#endif
+	.open   = mp_chr_open,
+	.release = mp_chr_close,
+};
+
+static struct miscdevice mp_miscdev = {
+	.minor = MISC_DYNAMIC_MINOR,
+	.name = "mp",
+	.nodename = "net/mp",
+	.fops = &mp_fops,
+};
+
+static int mp_device_event(struct notifier_block *unused,
+		unsigned long event, void *ptr)
+{
+	struct net_device *dev = ptr;
+	struct mpassthru_port *port;
+	struct mp_struct *mp = NULL;
+	struct socket *sock = NULL;
+	struct sock *sk;
+
+	port = dev->mp_port;
+	if (port == NULL)
+		return NOTIFY_DONE;
+
+	switch (event) {
+	case NETDEV_UNREGISTER:
+		sock = dev->mp_port->sock;
+		mp = container_of(sock->sk, struct mp_sock, sk)->mp;
+		do_unbind(mp->mfile);
+		break;
+	case NETDEV_CHANGE:
+		sk = dev->mp_port->sock->sk;
+		sk->sk_state_change(sk);
+		break;
+	}
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block mp_notifier_block __read_mostly = {
+	.notifier_call  = mp_device_event,
+};
+
+static int mp_init(void)
+{
+	int err = 0;
+
+	ext_page_info_cache = kmem_cache_create("skb_page_info",
+						sizeof(struct page_info),
+						0, SLAB_HWCACHE_ALIGN, NULL);
+	if (!ext_page_info_cache)
+		return -ENOMEM;
+
+	err = misc_register(&mp_miscdev);
+	if (err) {
+		printk(KERN_ERR "mp: Can't register misc device\n");
+		kmem_cache_destroy(ext_page_info_cache);
+	} else {
+		printk(KERN_INFO "Registering mp misc device - minor = %d\n",
+				mp_miscdev.minor);
+		register_netdevice_notifier(&mp_notifier_block);
+	}
+	return err;
+}
+
+void mp_exit(void)
+{
+	unregister_netdevice_notifier(&mp_notifier_block);
+	misc_deregister(&mp_miscdev);
+	kmem_cache_destroy(ext_page_info_cache);
+}
+
+/* Get an underlying socket object from mp file.  Returns error unless file is
+ * attached to a device.  The returned object works like a packet socket, it
+ * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
+ * holding a reference to the file for as long as the socket is in use. */
+struct socket *mp_get_socket(struct file *file)
+{
+	struct mp_file *mfile = file->private_data;
+	struct mp_struct *mp;
+
+	if (file->f_op != &mp_fops)
+		return ERR_PTR(-EINVAL);
+	mp = mp_get(mfile);
+	if (!mp)
+		return ERR_PTR(-EBADFD);
+	mp_put(mfile);
+	return &mp->socket;
+}
+EXPORT_SYMBOL_GPL(mp_get_socket);
+
+module_init(mp_init);
+module_exit(mp_exit);
+MODULE_AUTHOR(DRV_COPYRIGHT);
+MODULE_DESCRIPTION(DRV_DESCRIPTION);
+MODULE_LICENSE("GPL v2");
-- 
1.5.4.4

[RFC PATCH v10 13/16] Add mp(mediate passthru) device.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

The patch add mp(mediate passthru) device, which now
based on vhost-net backend driver and provides proto_ops
to send/receive guest buffers data from/to guest vitio-net
driver.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
 drivers/vhost/mpassthru.c | 1407 +++++++++++++++++++++++++++++++++++++++++++++
 1 files changed, 1407 insertions(+), 0 deletions(-)
 create mode 100644 drivers/vhost/mpassthru.c

diff --git a/drivers/vhost/mpassthru.c b/drivers/vhost/mpassthru.c
new file mode 100644
index 0000000..d86d94c
--- /dev/null
+++ b/drivers/vhost/mpassthru.c
@@ -0,0 +1,1407 @@
+/*
+ *  MPASSTHRU - Mediate passthrough device.
+ *  Copyright (C) 2009 ZhaoYu, XinXiaohui, Dike, Jeffery G
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ *  GNU General Public License for more details.
+ *
+ */
+
+#define DRV_NAME        "mpassthru"
+#define DRV_DESCRIPTION "Mediate passthru device driver"
+#define DRV_COPYRIGHT   "(C) 2009 ZhaoYu, XinXiaohui, Dike, Jeffery G"
+
+#include <linux/compat.h>
+#include <linux/module.h>
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/major.h>
+#include <linux/slab.h>
+#include <linux/smp_lock.h>
+#include <linux/poll.h>
+#include <linux/fcntl.h>
+#include <linux/init.h>
+#include <linux/aio.h>
+
+#include <linux/skbuff.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/miscdevice.h>
+#include <linux/ethtool.h>
+#include <linux/rtnetlink.h>
+#include <linux/if.h>
+#include <linux/if_arp.h>
+#include <linux/if_ether.h>
+#include <linux/crc32.h>
+#include <linux/nsproxy.h>
+#include <linux/uaccess.h>
+#include <linux/virtio_net.h>
+#include <linux/mpassthru.h>
+#include <net/net_namespace.h>
+#include <net/netns/generic.h>
+#include <net/rtnetlink.h>
+#include <net/sock.h>
+
+#include <asm/system.h>
+
+/* Uncomment to enable debugging */
+/* #define MPASSTHRU_DEBUG 1 */
+
+#ifdef MPASSTHRU_DEBUG
+static int debug;
+
+#define DBG  if (mp->debug) printk
+#define DBG1 if (debug == 2) printk
+#else
+#define DBG(a...)
+#define DBG1(a...)
+#endif
+
+#define COPY_THRESHOLD (L1_CACHE_BYTES * 4)
+#define COPY_HDR_LEN   (L1_CACHE_BYTES < 64 ? 64 : L1_CACHE_BYTES)
+
+struct frag {
+	u16     offset;
+	u16     size;
+};
+
+#define	HASH_BUCKETS	(8192*2)
+
+struct page_info {
+	struct list_head        list;
+	struct page_info	*next;
+	struct page_info	*prev;
+	struct page             *pages[MAX_SKB_FRAGS];
+	struct sk_buff          *skb;
+	struct page_ctor        *ctor;
+
+	/* The pointer relayed to skb, to indicate
+	 * it's a external allocated skb or kernel
+	 */
+	struct skb_ext_page    ext_page;
+
+#define INFO_READ                      0
+#define INFO_WRITE                     1
+	unsigned                flags;
+	unsigned                pnum;
+
+	/* The fields after that is for backend
+	 * driver, now for vhost-net.
+	 */
+
+	struct kiocb            *iocb;
+	unsigned int            desc_pos;
+	struct iovec            hdr[2];
+	struct iovec            iov[MAX_SKB_FRAGS];
+};
+
+static struct kmem_cache *ext_page_info_cache;
+
+struct page_ctor {
+	struct list_head        readq;
+	int			wq_len;
+	int			rq_len;
+	spinlock_t		read_lock;
+	/* record the locked pages */
+	int			lock_pages;
+	struct rlimit		o_rlim;
+	struct net_device	*dev;
+	struct mpassthru_port	port;
+	struct page_info	**hash_table;
+};
+
+struct mp_struct {
+	struct mp_file		*mfile;
+	struct net_device       *dev;
+	struct page_ctor	*ctor;
+	struct socket           socket;
+
+#ifdef MPASSTHRU_DEBUG
+	int debug;
+#endif
+};
+
+struct mp_file {
+	atomic_t count;
+	struct mp_struct *mp;
+	struct net *net;
+};
+
+struct mp_sock {
+	struct sock		sk;
+	struct mp_struct	*mp;
+};
+
+static int mp_dev_change_flags(struct net_device *dev, unsigned flags)
+{
+	int ret = 0;
+
+	rtnl_lock();
+	ret = dev_change_flags(dev, flags);
+	rtnl_unlock();
+
+	if (ret < 0)
+		printk(KERN_ERR "failed to change dev state of %s", dev->name);
+
+	return ret;
+}
+
+/* The main function to allocate external buffers */
+static struct skb_ext_page *page_ctor(struct mpassthru_port *port,
+		struct sk_buff *skb, int npages)
+{
+	int i;
+	unsigned long flags;
+	struct page_ctor *ctor;
+	struct page_info *info = NULL;
+
+	ctor = container_of(port, struct page_ctor, port);
+
+	spin_lock_irqsave(&ctor->read_lock, flags);
+	if (!list_empty(&ctor->readq)) {
+		info = list_first_entry(&ctor->readq, struct page_info, list);
+		list_del(&info->list);
+		ctor->rq_len--;
+	}
+	spin_unlock_irqrestore(&ctor->read_lock, flags);
+	if (!info)
+		return NULL;
+
+	for (i = 0; i < info->pnum; i++)
+		get_page(info->pages[i]);
+	info->skb = skb;
+	return &info->ext_page;
+}
+
+static struct page_info *mp_hash_lookup(struct page_ctor *ctor,
+					struct page *page);
+static struct page_info *mp_hash_delete(struct page_ctor *ctor,
+					struct page_info *info);
+
+static struct skb_ext_page *mp_lookup(struct net_device *dev,
+				      struct page *page)
+{
+	struct mp_struct *mp =
+		container_of(dev->mp_port->sock->sk, struct mp_sock, sk)->mp;
+	struct page_ctor *ctor = mp->ctor;
+	struct page_info *info;
+
+	info = mp_hash_lookup(ctor, page);
+	if (!info)
+		return NULL;
+	return &info->ext_page;
+}
+
+static int page_ctor_attach(struct mp_struct *mp)
+{
+	int rc;
+	struct page_ctor *ctor;
+	struct net_device *dev = mp->dev;
+
+	/* locked by mp_mutex */
+	if (mp->ctor)
+		return -EBUSY;
+
+	ctor = kzalloc(sizeof(*ctor), GFP_KERNEL);
+	if (!ctor)
+		return -ENOMEM;
+	rc = netdev_mp_port_prep(dev, &ctor->port);
+	if (rc)
+		goto fail;
+
+	INIT_LIST_HEAD(&ctor->readq);
+	spin_lock_init(&ctor->read_lock);
+	ctor->hash_table = kzalloc(sizeof(struct page_info *) * HASH_BUCKETS,
+			GFP_KERNEL);
+	if (!ctor->hash_table)
+		goto fail_hash;
+
+	ctor->rq_len = 0;
+	ctor->wq_len = 0;
+
+	dev_hold(dev);
+	ctor->dev = dev;
+	ctor->port.ctor = page_ctor;
+	ctor->port.sock = &mp->socket;
+	ctor->port.hash = mp_lookup;
+	ctor->lock_pages = 0;
+
+	/* locked by mp_mutex */
+	dev->mp_port = &ctor->port;
+	mp->ctor = ctor;
+
+	return 0;
+
+fail_hash:
+	kfree(ctor->hash_table);
+
+fail:
+	kfree(ctor);
+	dev_put(dev);
+
+	return rc;
+}
+
+struct page_info *info_dequeue(struct page_ctor *ctor)
+{
+	unsigned long flags;
+	struct page_info *info = NULL;
+	spin_lock_irqsave(&ctor->read_lock, flags);
+	if (!list_empty(&ctor->readq)) {
+		info = list_first_entry(&ctor->readq,
+				struct page_info, list);
+		list_del(&info->list);
+		ctor->rq_len--;
+	}
+	spin_unlock_irqrestore(&ctor->read_lock, flags);
+	return info;
+}
+
+static int set_memlock_rlimit(struct page_ctor *ctor, int resource,
+			      unsigned long cur, unsigned long max)
+{
+	struct rlimit new_rlim, *old_rlim;
+	int retval;
+
+	if (resource != RLIMIT_MEMLOCK)
+		return -EINVAL;
+	new_rlim.rlim_cur = cur;
+	new_rlim.rlim_max = max;
+
+	old_rlim = current->signal->rlim + resource;
+
+	/* remember the old rlimit value when backend enabled */
+	ctor->o_rlim.rlim_cur = old_rlim->rlim_cur;
+	ctor->o_rlim.rlim_max = old_rlim->rlim_max;
+
+	if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
+			!capable(CAP_SYS_RESOURCE))
+		return -EPERM;
+
+	retval = security_task_setrlimit(resource, &new_rlim);
+	if (retval)
+		return retval;
+
+	task_lock(current->group_leader);
+	*old_rlim = new_rlim;
+	task_unlock(current->group_leader);
+	return 0;
+}
+
+static void relinquish_resource(struct page_ctor *ctor)
+{
+	if (!(ctor->dev->flags & IFF_UP) &&
+			!(ctor->wq_len + ctor->rq_len))
+		printk(KERN_INFO "relinquish_resource\n");
+}
+
+static void mp_ki_dtor(struct kiocb *iocb)
+{
+	struct page_info *info = (struct page_info *)(iocb->private);
+	int i;
+
+	if (info->flags == INFO_READ) {
+		for (i = 0; i < info->pnum; i++) {
+			if (info->pages[i]) {
+				set_page_dirty_lock(info->pages[i]);
+				put_page(info->pages[i]);
+			}
+		}
+		mp_hash_delete(info->ctor, info);
+		if (info->skb) {
+			info->skb->destructor = NULL;
+			kfree_skb(info->skb);
+		}
+		info->ctor->rq_len--;
+	} else
+		info->ctor->wq_len--;
+	/* Decrement the number of locked pages */
+	info->ctor->lock_pages -= info->pnum;
+	kmem_cache_free(ext_page_info_cache, info);
+	relinquish_resource(info->ctor);
+
+	return;
+}
+
+static struct kiocb *create_iocb(struct page_info *info, int size)
+{
+	struct kiocb *iocb = NULL;
+
+	iocb = info->iocb;
+	if (!iocb)
+		return iocb;
+	iocb->ki_flags = 0;
+	iocb->ki_users = 1;
+	iocb->ki_key = 0;
+	iocb->ki_ctx = NULL;
+	iocb->ki_cancel = NULL;
+	iocb->ki_retry = NULL;
+	iocb->ki_eventfd = NULL;
+	iocb->ki_pos = info->desc_pos;
+	iocb->ki_nbytes = size;
+	iocb->ki_dtor(iocb);
+	iocb->private = (void *)info;
+	iocb->ki_dtor = mp_ki_dtor;
+
+	return iocb;
+}
+
+static int page_ctor_detach(struct mp_struct *mp)
+{
+	struct page_ctor *ctor;
+	struct page_info *info;
+	int i;
+
+	/* locked by mp_mutex */
+	ctor = mp->ctor;
+	if (!ctor)
+		return -ENODEV;
+
+	while ((info = info_dequeue(ctor))) {
+		for (i = 0; i < info->pnum; i++)
+			if (info->pages[i])
+				put_page(info->pages[i]);
+		create_iocb(info, 0);
+		ctor->rq_len--;
+		kmem_cache_free(ext_page_info_cache, info);
+	}
+
+	relinquish_resource(ctor);
+
+	set_memlock_rlimit(ctor, RLIMIT_MEMLOCK,
+			   ctor->o_rlim.rlim_cur,
+			   ctor->o_rlim.rlim_max);
+
+	/* locked by mp_mutex */
+	ctor->dev->mp_port = NULL;
+	dev_put(ctor->dev);
+
+	mp->ctor = NULL;
+	kfree(ctor->hash_table);
+	kfree(ctor);
+	return 0;
+}
+
+static void __mp_detach(struct mp_struct *mp)
+{
+	mp->mfile = NULL;
+
+	mp_dev_change_flags(mp->dev, mp->dev->flags & ~IFF_UP);
+	page_ctor_detach(mp);
+	mp_dev_change_flags(mp->dev, mp->dev->flags | IFF_UP);
+
+	/* Drop the extra count on the net device */
+	dev_put(mp->dev);
+}
+
+static DEFINE_MUTEX(mp_mutex);
+
+static void mp_detach(struct mp_struct *mp)
+{
+	mutex_lock(&mp_mutex);
+	__mp_detach(mp);
+	mutex_unlock(&mp_mutex);
+}
+
+static struct mp_struct *mp_get(struct mp_file *mfile)
+{
+	struct mp_struct *mp = NULL;
+	if (atomic_inc_not_zero(&mfile->count))
+		mp = mfile->mp;
+
+	return mp;
+}
+
+static void mp_put(struct mp_file *mfile)
+{
+	if (atomic_dec_and_test(&mfile->count))
+		mp_detach(mfile->mp);
+}
+
+static void iocb_tag(struct kiocb *iocb)
+{
+	iocb->ki_flags = 1;
+}
+
+/* The callback to destruct the external buffers or skb */
+static void page_dtor(struct skb_ext_page *ext_page)
+{
+	struct page_info *info;
+	struct page_ctor *ctor;
+	struct sock *sk;
+	struct sk_buff *skb;
+
+	if (!ext_page)
+		return;
+	info = container_of(ext_page, struct page_info, ext_page);
+	if (!info)
+		return;
+	ctor = info->ctor;
+	skb = info->skb;
+
+	if (info->flags == INFO_READ) {
+		create_iocb(info, 0);
+		return;
+	}
+
+	/* For transmit, we should wait for the DMA finish by hardware.
+	 * Queue the notifier to wake up the backend driver
+	 */
+
+	iocb_tag(info->iocb);
+	sk = ctor->port.sock->sk;
+	sk->sk_write_space(sk);
+
+	return;
+}
+
+/* For small exteranl buffers transmit, we don't need to call
+ * get_user_pages().
+ */
+static struct page_info *alloc_small_page_info(struct page_ctor *ctor,
+		struct kiocb *iocb, int total)
+{
+	struct page_info *info =
+		kmem_cache_alloc(ext_page_info_cache, GFP_KERNEL);
+
+	if (!info)
+		return NULL;
+	info->ext_page.dtor = page_dtor;
+	info->ctor = ctor;
+	info->flags = INFO_WRITE;
+	info->iocb = iocb;
+	info->pnum = 0;
+	return info;
+}
+
+typedef u32 key_mp_t;
+static inline key_mp_t mp_hash(struct page *page, int buckets)
+{
+	key_mp_t k;
+
+	k = ((((unsigned long)page << 32UL) >> 32UL) / 0x38) % buckets ;
+	return k;
+}
+
+static void mp_hash_insert(struct page_ctor *ctor,
+		struct page *page, struct page_info *page_info)
+{
+	struct page_info *tmp;
+	key_mp_t key = mp_hash(page, HASH_BUCKETS);
+	if (!ctor->hash_table[key]) {
+		ctor->hash_table[key] = page_info;
+		return;
+	}
+
+	tmp = ctor->hash_table[key];
+	while (tmp->next)
+		tmp = tmp->next;
+
+	tmp->next = page_info;
+	page_info->prev = tmp;
+	return;
+}
+
+static struct page_info *mp_hash_delete(struct page_ctor *ctor,
+					struct page_info *info)
+{
+	key_mp_t key = mp_hash(info->pages[0], HASH_BUCKETS);
+	struct page_info *tmp = NULL;
+	int i;
+
+	tmp = ctor->hash_table[key];
+	while (tmp) {
+		if (tmp == info) {
+			if (!tmp->prev) {
+				ctor->hash_table[key] = tmp->next;
+				if (tmp->next)
+					tmp->next->prev = NULL;
+			} else {
+				tmp->prev->next = tmp->next;
+				if (tmp->next)
+					tmp->next->prev = tmp->prev;
+			}
+			return tmp;
+		}
+		tmp = tmp->next;
+	}
+	return tmp;
+}
+
+static struct page_info *mp_hash_lookup(struct page_ctor *ctor,
+					struct page *page)
+{
+	key_mp_t key = mp_hash(page, HASH_BUCKETS);
+	struct page_info *tmp = NULL;
+
+	int i;
+	tmp = ctor->hash_table[key];
+	while (tmp) {
+		for (i = 0; i < tmp->pnum; i++) {
+			if (tmp->pages[i] == page)
+				return tmp;
+		}
+		tmp = tmp->next;
+	}
+	return tmp;
+}
+
+/* The main function to transform the guest user space address
+ * to host kernel address via get_user_pages(). Thus the hardware
+ * can do DMA directly to the external buffer address.
+ */
+static struct page_info *alloc_page_info(struct page_ctor *ctor,
+		struct kiocb *iocb, struct iovec *iov,
+		int count, struct frag *frags,
+		int npages, int total)
+{
+	int rc;
+	int i, j, n = 0;
+	int len;
+	unsigned long base, lock_limit;
+	struct page_info *info = NULL;
+
+	lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur;
+	lock_limit >>= PAGE_SHIFT;
+
+	if (ctor->lock_pages + count > lock_limit && npages) {
+		printk(KERN_INFO "exceed the locked memory rlimit.");
+		return NULL;
+	}
+
+	info = kmem_cache_alloc(ext_page_info_cache, GFP_KERNEL);
+	
+	if (!info)
+		return NULL;
+	info->skb = NULL;
+	info->next = info->prev = NULL;
+
+	for (i = j = 0; i < count; i++) {
+		base = (unsigned long)iov[i].iov_base;
+		len = iov[i].iov_len;
+
+		if (!len)
+			continue;
+		n = ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
+
+		rc = get_user_pages_fast(base, n, npages ? 1 : 0,
+				&info->pages[j]);
+		if (rc != n)
+			goto failed;
+
+		while (n--) {
+			frags[j].offset = base & ~PAGE_MASK;
+			frags[j].size = min_t(int, len,
+					PAGE_SIZE - frags[j].offset);
+			len -= frags[j].size;
+			base += frags[j].size;
+			j++;
+		}
+	}
+
+#ifdef CONFIG_HIGHMEM
+	if (npages && !(dev->features & NETIF_F_HIGHDMA)) {
+		for (i = 0; i < j; i++) {
+			if (PageHighMem(info->pages[i]))
+				goto failed;
+		}
+	}
+#endif
+
+	info->ext_page.dtor = page_dtor;
+	info->ext_page.page = info->pages[0];
+	info->ctor = ctor;
+	info->pnum = j;
+	info->iocb = iocb;
+	if (!npages)
+		info->flags = INFO_WRITE;
+	else
+		info->flags = INFO_READ;
+
+	if (info->flags == INFO_READ) {
+		if (frags[0].offset == 0 && iocb->ki_iovec[0].iov_len) {
+			frags[0].offset = iocb->ki_iovec[0].iov_len;
+			ctor->port.vnet_hlen = iocb->ki_iovec[0].iov_len;
+		}
+		for (i = 0; i < j; i++)
+			mp_hash_insert(ctor, info->pages[i], info);
+	}
+	/* increment the number of locked pages */
+	ctor->lock_pages += j;
+	return info;
+
+failed:
+	for (i = 0; i < j; i++)
+		put_page(info->pages[i]);
+
+	kmem_cache_free(ext_page_info_cache, info);
+
+	return NULL;
+}
+
+static void mp_sock_destruct(struct sock *sk)
+{
+	struct mp_struct *mp = container_of(sk, struct mp_sock, sk)->mp;
+	kfree(mp);
+}
+
+static void mp_sock_state_change(struct sock *sk)
+{
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN);
+}
+
+static void mp_sock_write_space(struct sock *sk)
+{
+	if (sk_has_sleeper(sk))
+		wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT);
+}
+
+static void mp_sock_data_ready(struct sock *sk, int coming)
+{
+	struct mp_struct *mp = container_of(sk, struct mp_sock, sk)->mp;
+	struct page_ctor *ctor = NULL;
+	struct sk_buff *skb = NULL;
+	struct page_info *info = NULL;
+	int len;
+
+	ctor = mp->ctor;
+	if (!ctor)
+		return;
+
+	while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
+		struct page *page;
+		int off;
+		int size = 0, i = 0;
+		struct skb_shared_info *shinfo = skb_shinfo(skb);
+		struct skb_ext_page *ext_page =
+			(struct skb_ext_page *)(shinfo->destructor_arg);
+		struct virtio_net_hdr_mrg_rxbuf hdr = {
+			.hdr.flags = 0,
+			.hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE
+		};
+
+		if (skb->ip_summed == CHECKSUM_COMPLETE)
+			DBG(KERN_INFO "csum %d\n", skb->ip_summed);
+
+		if (shinfo->frags[0].page == ext_page->page) {
+			info = container_of(ext_page,
+					    struct page_info,
+					    ext_page);
+			if (shinfo->nr_frags)
+				hdr.num_buffers = shinfo->nr_frags;
+			else
+				hdr.num_buffers = shinfo->nr_frags + 1;
+		} else {
+			info = container_of(ext_page,
+					    struct page_info,
+					    ext_page);
+			hdr.num_buffers = shinfo->nr_frags + 1;
+		}
+		skb_push(skb, ETH_HLEN);
+
+		if (skb_is_gso(skb)) {
+			hdr.hdr.hdr_len = skb_headlen(skb);
+			hdr.hdr.gso_size = shinfo->gso_size;
+			if (shinfo->gso_type & SKB_GSO_TCPV4)
+				hdr.hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+			else if (shinfo->gso_type & SKB_GSO_TCPV6)
+				hdr.hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
+			else if (shinfo->gso_type & SKB_GSO_UDP)
+				hdr.hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
+			else
+				BUG();
+			if (shinfo->gso_type & SKB_GSO_TCP_ECN)
+				hdr.hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
+
+		} else
+			hdr.hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
+
+		if (skb->ip_summed == CHECKSUM_PARTIAL) {
+			hdr.hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
+			hdr.hdr.csum_start =
+				skb->csum_start - skb_headroom(skb);
+			hdr.hdr.csum_offset = skb->csum_offset;
+		}
+
+		off = info->hdr[0].iov_len;
+		len = memcpy_toiovec(info->iov, (unsigned char *)&hdr, off);
+		if (len) {
+			DBG(KERN_INFO
+				"Unable to write vnet_hdr at addr %p: %d\n",
+				info->iov_base, len);
+			goto clean;
+		}
+
+		memcpy_toiovec(info->iov, skb->data, skb_headlen(skb));
+
+		info->iocb->ki_left = hdr.num_buffers;
+		if (shinfo->frags[0].page == ext_page->page) {
+			size = shinfo->frags[0].size +
+				shinfo->frags[0].page_offset - off;
+			i = 1;
+		} else {
+			size = skb_headlen(skb);
+			i = 0;
+		}
+		create_iocb(info, off + size);
+		for (i = i; i < shinfo->nr_frags; i++) {
+			page = shinfo->frags[i].page;
+			info = mp_hash_lookup(ctor, shinfo->frags[i].page);
+			create_iocb(info, shinfo->frags[i].size);
+		}
+		info->skb = skb;
+		shinfo->nr_frags = 0;
+		shinfo->destructor_arg = NULL;
+		continue;
+clean:
+		kfree_skb(skb);
+		for (i = 0; i < info->pnum; i++)
+			put_page(info->pages[i]);
+		kmem_cache_free(ext_page_info_cache, info);
+	}
+	return;
+}
+
+static inline struct sk_buff *mp_alloc_skb(struct sock *sk, size_t prepad,
+					   size_t len, size_t linear,
+					   int noblock, int *err)
+{
+	struct sk_buff *skb;
+
+	/* Under a page?  Don't bother with paged skb. */
+	if (prepad + len < PAGE_SIZE || !linear)
+		linear = len;
+
+	skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
+			err);
+	if (!skb)
+		return NULL;
+
+	skb_reserve(skb, prepad);
+	skb_put(skb, linear);
+	skb->data_len = len - linear;
+	skb->len += len - linear;
+
+	return skb;
+}
+
+static int mp_skb_from_vnet_hdr(struct sk_buff *skb,
+		struct virtio_net_hdr *vnet_hdr)
+{
+	unsigned short gso_type = 0;
+	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+		switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
+		case VIRTIO_NET_HDR_GSO_TCPV4:
+			gso_type = SKB_GSO_TCPV4;
+			break;
+		case VIRTIO_NET_HDR_GSO_TCPV6:
+			gso_type = SKB_GSO_TCPV6;
+			break;
+		case VIRTIO_NET_HDR_GSO_UDP:
+			gso_type = SKB_GSO_UDP;
+			break;
+		default:
+			return -EINVAL;
+		}
+
+		if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
+			gso_type |= SKB_GSO_TCP_ECN;
+
+		if (vnet_hdr->gso_size == 0)
+			return -EINVAL;
+	}
+
+	if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
+		if (!skb_partial_csum_set(skb, vnet_hdr->csum_start,
+					vnet_hdr->csum_offset))
+			return -EINVAL;
+	}
+
+	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+		skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
+		skb_shinfo(skb)->gso_type = gso_type;
+
+		/* Header must be checked, and gso_segs computed. */
+		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
+		skb_shinfo(skb)->gso_segs = 0;
+	}
+	return 0;
+}
+
+static int mp_sendmsg(struct kiocb *iocb, struct socket *sock,
+		struct msghdr *m, size_t total_len)
+{
+	struct mp_struct *mp = container_of(sock->sk, struct mp_sock, sk)->mp;
+	struct virtio_net_hdr vnet_hdr = {0};
+	int hdr_len = 0;
+	struct page_ctor *ctor;
+	struct iovec *iov = m->msg_iov;
+	struct page_info *info = NULL;
+	struct frag frags[MAX_SKB_FRAGS];
+	struct sk_buff *skb;
+	int count = m->msg_iovlen;
+	int total = 0, header, n, i, len, rc;
+	unsigned long base;
+
+	ctor = mp->ctor;
+	if (!ctor)
+		return -ENODEV;
+
+	total = iov_length(iov, count);
+
+	if (total < ETH_HLEN)
+		return -EINVAL;
+
+	if (total <= COPY_THRESHOLD)
+		goto copy;
+
+	n = 0;
+	for (i = 0; i < count; i++) {
+		base = (unsigned long)iov[i].iov_base;
+		len = iov[i].iov_len;
+		if (!len)
+			continue;
+		n += ((base & ~PAGE_MASK) + len + ~PAGE_MASK) >> PAGE_SHIFT;
+		if (n > MAX_SKB_FRAGS)
+			return -EINVAL;
+	}
+
+copy:
+	hdr_len = sizeof(vnet_hdr);
+	if ((total - iocb->ki_iovec[0].iov_len) < 0)
+		return -EINVAL;
+
+	rc = memcpy_fromiovecend((void *)&vnet_hdr, iocb->ki_iovec, 0, hdr_len);
+	if (rc < 0)
+		return -EINVAL;
+
+	if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
+			vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
+			vnet_hdr.hdr_len)
+		vnet_hdr.hdr_len = vnet_hdr.csum_start +
+			vnet_hdr.csum_offset + 2;
+
+	if (vnet_hdr.hdr_len > total)
+		return -EINVAL;
+
+	header = total > COPY_THRESHOLD ? COPY_HDR_LEN : total;
+
+	skb = mp_alloc_skb(sock->sk, NET_IP_ALIGN, header,
+			   iocb->ki_iovec[0].iov_len, 1, &rc);
+
+	if (!skb)
+		goto drop;
+
+	skb_set_network_header(skb, ETH_HLEN);
+	memcpy_fromiovec(skb->data, iov, header);
+
+	skb_reset_mac_header(skb);
+	skb->protocol = eth_hdr(skb)->h_proto;
+
+	rc = mp_skb_from_vnet_hdr(skb, &vnet_hdr);
+	if (rc)
+		goto drop;
+
+	if (header == total) {
+		rc = total;
+		info = alloc_small_page_info(ctor, iocb, total);
+	} else {
+		info = alloc_page_info(ctor, iocb, iov, count, frags, 0, total);
+		if (info)
+			for (i = 0; i < info->pnum; i++) {
+				skb_add_rx_frag(skb, i, info->pages[i],
+						frags[i].offset, frags[i].size);
+				info->pages[i] = NULL;
+			}
+	}
+	if (info != NULL) {
+		info->desc_pos = iocb->ki_pos;
+		info->skb = skb;
+		skb_shinfo(skb)->destructor_arg = &info->ext_page;
+		skb->dev = mp->dev;
+		ctor->wq_len++;
+		create_iocb(info, total);
+		dev_queue_xmit(skb);
+		if (!ctor->rq_len)
+			sock->sk->sk_state_change(sock->sk);
+		return 0;
+	}
+drop:
+	kfree_skb(skb);
+	if (info) {
+		for (i = 0; i < info->pnum; i++)
+			put_page(info->pages[i]);
+		kmem_cache_free(ext_page_info_cache, info);
+	}
+	mp->dev->stats.tx_dropped++;
+	return -ENOMEM;
+}
+
+static int mp_recvmsg(struct kiocb *iocb, struct socket *sock,
+		struct msghdr *m, size_t total_len,
+		int flags)
+{
+	struct mp_struct *mp = container_of(sock->sk, struct mp_sock, sk)->mp;
+	struct page_ctor *ctor;
+	struct iovec *iov = m->msg_iov;
+	int count = m->msg_iovlen;
+	int npages, payload;
+	struct page_info *info;
+	struct frag frags[MAX_SKB_FRAGS];
+	unsigned long base;
+	int i, len;
+	unsigned long flag;
+
+	if (!(flags & MSG_DONTWAIT))
+		return -EINVAL;
+
+	ctor = mp->ctor;
+	if (!ctor)
+		return -EINVAL;
+
+	/* Error detections in case invalid external buffer */
+	if (count > 2 && iov[1].iov_len < ctor->port.hdr_len &&
+			mp->dev->features & NETIF_F_SG) {
+		return -EINVAL;
+	}
+
+	npages = ctor->port.npages;
+	payload = ctor->port.data_len;
+
+	/* If KVM guest virtio-net FE driver use SG feature */
+	if (count > 2) {
+		for (i = 2; i < count; i++) {
+			base = (unsigned long)iov[i].iov_base & ~PAGE_MASK;
+			len = iov[i].iov_len;
+			if (npages == 1)
+				len = min_t(int, len, PAGE_SIZE - base);
+			else if (base)
+				break;
+			payload -= len;
+			if (payload <= 0)
+				goto proceed;
+			if (npages == 1 || (len & ~PAGE_MASK))
+				break;
+		}
+	}
+
+	if ((((unsigned long)iov[1].iov_base & ~PAGE_MASK)
+				- NET_SKB_PAD - NET_IP_ALIGN) >= 0)
+		goto proceed;
+
+	return -EINVAL;
+
+proceed:
+	/* skip the virtnet head */
+	if (count > 1) {
+		iov++;
+		count--;
+	}
+
+	if (!ctor->lock_pages || !ctor->rq_len) {
+		set_memlock_rlimit(ctor, RLIMIT_MEMLOCK,
+				iocb->ki_user_data * 4096 * 2,
+				iocb->ki_user_data * 4096 * 2);
+	}
+
+	/* Translate address to kernel */
+	info = alloc_page_info(ctor, iocb, iov, count, frags, npages, 0);
+	if (!info)
+		return -ENOMEM;
+	info->hdr[0].iov_base = iocb->ki_iovec[0].iov_base;
+	info->hdr[0].iov_len = iocb->ki_iovec[0].iov_len;
+	iocb->ki_iovec[0].iov_len = 0;
+	iocb->ki_left = 0;
+	info->desc_pos = iocb->ki_pos;
+
+	if (count > 1) {
+		iov--;
+		count++;
+	}
+
+	memcpy(info->iov, iov, sizeof(struct iovec) * count);
+
+	spin_lock_irqsave(&ctor->read_lock, flag);
+	list_add_tail(&info->list, &ctor->readq);
+	spin_unlock_irqrestore(&ctor->read_lock, flag);
+
+	ctor->rq_len++;
+
+	return 0;
+}
+
+/* Ops structure to mimic raw sockets with mp device */
+static const struct proto_ops mp_socket_ops = {
+	.sendmsg = mp_sendmsg,
+	.recvmsg = mp_recvmsg,
+};
+
+static struct proto mp_proto = {
+	.name           = "mp",
+	.owner          = THIS_MODULE,
+	.obj_size       = sizeof(struct mp_sock),
+};
+
+static int mp_chr_open(struct inode *inode, struct file * file)
+{
+	struct mp_file *mfile;
+	cycle_kernel_lock();
+	DBG1(KERN_INFO "mp: mp_chr_open\n");
+
+	mfile = kzalloc(sizeof(*mfile), GFP_KERNEL);
+	if (!mfile)
+		return -ENOMEM;
+	atomic_set(&mfile->count, 0);
+	mfile->mp = NULL;
+	mfile->net = get_net(current->nsproxy->net_ns);
+	file->private_data = mfile;
+	return 0;
+}
+
+static int mp_attach(struct mp_struct *mp, struct file *file)
+{
+	struct mp_file *mfile = file->private_data;
+	int err;
+
+	netif_tx_lock_bh(mp->dev);
+
+	err = -EINVAL;
+
+	if (mfile->mp)
+		goto out;
+
+	err = -EBUSY;
+	if (mp->mfile)
+		goto out;
+
+	err = 0;
+	mfile->mp = mp;
+	mp->mfile = mfile;
+	mp->socket.file = file;
+	dev_hold(mp->dev);
+	sock_hold(mp->socket.sk);
+	atomic_inc(&mfile->count);
+
+out:
+	netif_tx_unlock_bh(mp->dev);
+	return err;
+}
+
+static int do_unbind(struct mp_file *mfile)
+{
+	struct mp_struct *mp = mp_get(mfile);
+
+	if (!mp)
+		return -EINVAL;
+
+	mp_detach(mp);
+	sock_put(mp->socket.sk);
+	mp_put(mfile);
+	return 0;
+}
+
+static long mp_chr_ioctl(struct file *file, unsigned int cmd,
+		unsigned long arg)
+{
+	struct mp_file *mfile = file->private_data;
+	struct mp_struct *mp;
+	struct net_device *dev;
+	void __user* argp = (void __user *)arg;
+	struct ifreq ifr;
+	struct sock *sk;
+	int ret;
+
+	ret = -EINVAL;
+
+	switch (cmd) {
+	case MPASSTHRU_BINDDEV:
+		ret = -EFAULT;
+		if (copy_from_user(&ifr, argp, sizeof ifr))
+			break;
+
+		ifr.ifr_name[IFNAMSIZ-1] = '\0';
+
+		ret = -ENODEV;
+		dev = dev_get_by_name(mfile->net, ifr.ifr_name);
+		if (!dev)
+			break;
+
+		mutex_lock(&mp_mutex);
+
+		ret = -EBUSY;
+
+		/* the device can be only bind once */
+		if (dev_is_mpassthru(dev))
+			goto err_dev_put;
+
+		mp = mfile->mp;
+		if (mp)
+			goto err_dev_put;
+
+		mp = kzalloc(sizeof(*mp), GFP_KERNEL);
+		if (!mp) {
+			ret = -ENOMEM;
+			goto err_dev_put;
+		}
+		mp->dev = dev;
+		ret = -ENOMEM;
+
+		sk = sk_alloc(mfile->net, AF_UNSPEC, GFP_KERNEL, &mp_proto);
+		if (!sk)
+			goto err_free_mp;
+
+		init_waitqueue_head(&mp->socket.wait);
+		mp->socket.ops = &mp_socket_ops;
+		sock_init_data(&mp->socket, sk);
+		sk->sk_sndbuf = INT_MAX;
+		container_of(sk, struct mp_sock, sk)->mp = mp;
+
+		sk->sk_destruct = mp_sock_destruct;
+		sk->sk_data_ready = mp_sock_data_ready;
+		sk->sk_write_space = mp_sock_write_space;
+		sk->sk_state_change = mp_sock_state_change;
+		ret = mp_attach(mp, file);
+		if (ret < 0)
+			goto err_free_sk;
+
+		ret = page_ctor_attach(mp);
+		if (ret < 0)
+			goto err_free_sk;
+		mp_dev_change_flags(mp->dev, mp->dev->flags & (~IFF_UP));
+		mp_dev_change_flags(mp->dev, mp->dev->flags | IFF_UP);
+		sk->sk_state_change(sk);
+out:
+		mutex_unlock(&mp_mutex);
+		break;
+err_free_sk:
+		sk_free(sk);
+err_free_mp:
+		kfree(mp);
+err_dev_put:
+		dev_put(dev);
+		goto out;
+
+	case MPASSTHRU_UNBINDDEV:
+		ret = do_unbind(mfile);
+		break;
+
+	default:
+		break;
+	}
+	return ret;
+}
+
+static unsigned int mp_chr_poll(struct file *file, poll_table * wait)
+{
+	struct mp_file *mfile = file->private_data;
+	struct mp_struct *mp = mp_get(mfile);
+	struct sock *sk;
+	unsigned int mask = 0;
+
+	if (!mp)
+		return POLLERR;
+
+	sk = mp->socket.sk;
+
+	poll_wait(file, &mp->socket.wait, wait);
+
+	if (!skb_queue_empty(&sk->sk_receive_queue))
+		mask |= POLLIN | POLLRDNORM;
+
+	if (sock_writeable(sk) ||
+		(!test_and_set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags) &&
+			 sock_writeable(sk)))
+		mask |= POLLOUT | POLLWRNORM;
+
+	if (mp->dev->reg_state != NETREG_REGISTERED)
+		mask = POLLERR;
+
+	mp_put(mfile);
+	return mask;
+}
+
+static ssize_t mp_chr_aio_write(struct kiocb *iocb, const struct iovec *iov,
+				unsigned long count, loff_t pos)
+{
+	struct file *file = iocb->ki_filp;
+	struct mp_struct *mp = mp_get(file->private_data);
+	struct sock *sk = mp->socket.sk;
+	struct sk_buff *skb;
+	int len, err;
+	ssize_t result = 0;
+
+	if (!mp)
+		return -EBADFD;
+
+	/* currently, async is not supported.
+	 * but we may support real async aio from user application,
+	 * maybe qemu virtio-net backend.
+	 */
+	if (!is_sync_kiocb(iocb))
+		return -EFAULT;
+
+	len = iov_length(iov, count);
+
+	if (unlikely(len) < ETH_HLEN)
+		return -EINVAL;
+
+	skb = sock_alloc_send_skb(sk, len + NET_IP_ALIGN,
+				  file->f_flags & O_NONBLOCK, &err);
+
+	if (!skb)
+		return -EFAULT;
+
+	skb_reserve(skb, NET_IP_ALIGN);
+	skb_put(skb, len);
+
+	if (skb_copy_datagram_from_iovec(skb, 0, iov, 0, len)) {
+		kfree_skb(skb);
+		return -EAGAIN;
+	}
+
+	skb->protocol = eth_type_trans(skb, mp->dev);
+	skb->dev = mp->dev;
+
+	dev_queue_xmit(skb);
+
+	mp_put(file->private_data);
+	return result;
+}
+
+static int mp_chr_close(struct inode *inode, struct file *file)
+{
+	struct mp_file *mfile = file->private_data;
+
+	/*
+	 * Ignore return value since an error only means there was nothing to
+	 * do
+	 */
+	do_unbind(mfile);
+
+	put_net(mfile->net);
+	kfree(mfile);
+
+	return 0;
+}
+
+#ifdef CONFIG_COMPAT
+static long mp_chr_compat_ioctl(struct file *f, unsigned int ioctl,
+				unsigned long arg)
+{
+	return mp_chr_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
+}
+#endif
+
+static const struct file_operations mp_fops = {
+	.owner  = THIS_MODULE,
+	.llseek = no_llseek,
+	.write  = do_sync_write,
+	.aio_write = mp_chr_aio_write,
+	.poll   = mp_chr_poll,
+	.unlocked_ioctl = mp_chr_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl = mp_chr_compat_ioctl,
+#endif
+	.open   = mp_chr_open,
+	.release = mp_chr_close,
+};
+
+static struct miscdevice mp_miscdev = {
+	.minor = MISC_DYNAMIC_MINOR,
+	.name = "mp",
+	.nodename = "net/mp",
+	.fops = &mp_fops,
+};
+
+static int mp_device_event(struct notifier_block *unused,
+		unsigned long event, void *ptr)
+{
+	struct net_device *dev = ptr;
+	struct mpassthru_port *port;
+	struct mp_struct *mp = NULL;
+	struct socket *sock = NULL;
+	struct sock *sk;
+
+	port = dev->mp_port;
+	if (port == NULL)
+		return NOTIFY_DONE;
+
+	switch (event) {
+	case NETDEV_UNREGISTER:
+		sock = dev->mp_port->sock;
+		mp = container_of(sock->sk, struct mp_sock, sk)->mp;
+		do_unbind(mp->mfile);
+		break;
+	case NETDEV_CHANGE:
+		sk = dev->mp_port->sock->sk;
+		sk->sk_state_change(sk);
+		break;
+	}
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block mp_notifier_block __read_mostly = {
+	.notifier_call  = mp_device_event,
+};
+
+static int mp_init(void)
+{
+	int err = 0;
+
+	ext_page_info_cache = kmem_cache_create("skb_page_info",
+						sizeof(struct page_info),
+						0, SLAB_HWCACHE_ALIGN, NULL);
+	if (!ext_page_info_cache)
+		return -ENOMEM;
+
+	err = misc_register(&mp_miscdev);
+	if (err) {
+		printk(KERN_ERR "mp: Can't register misc device\n");
+		kmem_cache_destroy(ext_page_info_cache);
+	} else {
+		printk(KERN_INFO "Registering mp misc device - minor = %d\n",
+				mp_miscdev.minor);
+		register_netdevice_notifier(&mp_notifier_block);
+	}
+	return err;
+}
+
+void mp_exit(void)
+{
+	unregister_netdevice_notifier(&mp_notifier_block);
+	misc_deregister(&mp_miscdev);
+	kmem_cache_destroy(ext_page_info_cache);
+}
+
+/* Get an underlying socket object from mp file.  Returns error unless file is
+ * attached to a device.  The returned object works like a packet socket, it
+ * can be used for sock_sendmsg/sock_recvmsg.  The caller is responsible for
+ * holding a reference to the file for as long as the socket is in use. */
+struct socket *mp_get_socket(struct file *file)
+{
+	struct mp_file *mfile = file->private_data;
+	struct mp_struct *mp;
+
+	if (file->f_op != &mp_fops)
+		return ERR_PTR(-EINVAL);
+	mp = mp_get(mfile);
+	if (!mp)
+		return ERR_PTR(-EBADFD);
+	mp_put(mfile);
+	return &mp->socket;
+}
+EXPORT_SYMBOL_GPL(mp_get_socket);
+
+module_init(mp_init);
+module_exit(mp_exit);
+MODULE_AUTHOR(DRV_COPYRIGHT);
+MODULE_DESCRIPTION(DRV_DESCRIPTION);
+MODULE_LICENSE("GPL v2");
-- 
1.5.4.4

[RFC PATCH v10 14/16] Provides multiple submits and asynchronous notifications.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

The vhost-net backend now only supports synchronous send/recv
operations. The patch provides multiple submits and asynchronous
notifications. This is needed for zero-copy case.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
---
 drivers/vhost/net.c   |  348 +++++++++++++++++++++++++++++++++++++++++++++----
 drivers/vhost/vhost.c |   79 +++++++++++
 drivers/vhost/vhost.h |   15 ++
 3 files changed, 414 insertions(+), 28 deletions(-)

diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index b38abc6..c4bc815 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -24,6 +24,8 @@
 #include <linux/if_arp.h>
 #include <linux/if_tun.h>
 #include <linux/if_macvlan.h>
+#include <linux/mpassthru.h>
+#include <linux/aio.h>
 
 #include <net/sock.h>
 
@@ -39,6 +41,8 @@ enum {
 	VHOST_NET_VQ_MAX = 2,
 };
 
+static struct kmem_cache *notify_cache;
+
 enum vhost_net_poll_state {
 	VHOST_NET_POLL_DISABLED = 0,
 	VHOST_NET_POLL_STARTED = 1,
@@ -49,6 +53,7 @@ struct vhost_net {
 	struct vhost_dev dev;
 	struct vhost_virtqueue vqs[VHOST_NET_VQ_MAX];
 	struct vhost_poll poll[VHOST_NET_VQ_MAX];
+	struct kmem_cache       *cache;
 	/* Tells us whether we are polling a socket for TX.
 	 * We only do this when socket buffer fills up.
 	 * Protected by tx vq lock. */
@@ -93,11 +98,190 @@ static void tx_poll_start(struct vhost_net *net, struct socket *sock)
 	net->tx_poll_state = VHOST_NET_POLL_STARTED;
 }
 
+struct kiocb *notify_dequeue(struct vhost_virtqueue *vq)
+{
+	struct kiocb *iocb = NULL;
+	unsigned long flags;
+
+	spin_lock_irqsave(&vq->notify_lock, flags);
+	if (!list_empty(&vq->notifier)) {
+		iocb = list_first_entry(&vq->notifier,
+				struct kiocb, ki_list);
+		list_del(&iocb->ki_list);
+	}
+	spin_unlock_irqrestore(&vq->notify_lock, flags);
+	return iocb;
+}
+
+static void handle_iocb(struct kiocb *iocb)
+{
+	struct vhost_virtqueue *vq = iocb->private;
+	unsigned long flags;
+
+	spin_lock_irqsave(&vq->notify_lock, flags);
+	list_add_tail(&iocb->ki_list, &vq->notifier);
+	spin_unlock_irqrestore(&vq->notify_lock, flags);
+}
+
+static int is_async_vq(struct vhost_virtqueue *vq)
+{
+	return (vq->link_state == VHOST_VQ_LINK_ASYNC);
+}
+
+static void handle_async_rx_events_notify(struct vhost_net *net,
+					  struct vhost_virtqueue *vq,
+					  struct socket *sock)
+{
+	struct kiocb *iocb = NULL;
+	struct vhost_log *vq_log = NULL;
+	int rx_total_len = 0;
+	unsigned int head, log, in, out;
+	int size;
+	int count;
+
+	struct virtio_net_hdr_mrg_rxbuf hdr = {
+		.hdr.flags = 0,
+		.hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE
+	};
+
+	if (!is_async_vq(vq))
+		return;
+
+	if (sock->sk->sk_data_ready)
+		sock->sk->sk_data_ready(sock->sk, 0);
+
+	vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ?
+		vq->log : NULL;
+
+	while ((iocb = notify_dequeue(vq)) != NULL) {
+		if (!iocb->ki_left) {
+			vhost_add_used_and_signal(&net->dev, vq,
+					iocb->ki_pos, iocb->ki_nbytes);
+			size = iocb->ki_nbytes;
+			head = iocb->ki_pos;
+			rx_total_len += iocb->ki_nbytes;
+
+			if (iocb->ki_dtor)
+				iocb->ki_dtor(iocb);
+			kmem_cache_free(net->cache, iocb);
+
+			/* when log is enabled, recomputing the log is needed,
+			 * since these buffers are in async queue, may not get
+			 * the log info before.
+			 */
+			if (unlikely(vq_log)) {
+				if (!log)
+					__vhost_get_desc(&net->dev, vq, vq->iov,
+							ARRAY_SIZE(vq->iov),
+							&out, &in, vq_log,
+							&log, head);
+				vhost_log_write(vq, vq_log, log, size);
+			}
+			if (unlikely(rx_total_len >= VHOST_NET_WEIGHT)) {
+				vhost_poll_queue(&vq->poll);
+				break;
+			}
+		} else {
+			int i = 0;
+			int count = iocb->ki_left;
+			int hc = count;
+			while (count--) {
+				if (iocb) {
+					vq->heads[i].id = iocb->ki_pos;
+					vq->heads[i].len = iocb->ki_nbytes;
+					size = iocb->ki_nbytes;
+					head = iocb->ki_pos;
+					rx_total_len += iocb->ki_nbytes;
+
+					if (iocb->ki_dtor)
+						iocb->ki_dtor(iocb);
+					kmem_cache_free(net->cache, iocb);
+
+					if (unlikely(vq_log)) {
+						if (!log)
+							__vhost_get_desc(
+							&net->dev, vq, vq->iov,
+							ARRAY_SIZE(vq->iov),
+							&out, &in, vq_log,
+							&log, head);
+						vhost_log_write(
+							vq, vq_log, log, size);
+					}
+				} else
+					break;
+
+				i++;
+				iocb == NULL;
+				if (count)
+					iocb = notify_dequeue(vq);
+			}
+			vhost_add_used_and_signal_n(
+					&net->dev, vq, vq->heads, hc);
+		}
+	}
+}
+
+static void handle_async_tx_events_notify(struct vhost_net *net,
+					  struct vhost_virtqueue *vq)
+{
+	struct kiocb *iocb = NULL;
+	struct list_head *entry, *tmp;
+	unsigned long flags;
+	int tx_total_len = 0;
+
+	if (!is_async_vq(vq))
+		return;
+
+	spin_lock_irqsave(&vq->notify_lock, flags);
+	list_for_each_safe(entry, tmp, &vq->notifier) {
+		iocb = list_entry(entry,
+				  struct kiocb, ki_list);
+		if (!iocb->ki_flags)
+			continue;
+		list_del(&iocb->ki_list);
+		vhost_add_used_and_signal(&net->dev, vq,
+				iocb->ki_pos, 0);
+		tx_total_len += iocb->ki_nbytes;
+
+		if (iocb->ki_dtor)
+			iocb->ki_dtor(iocb);
+
+		kmem_cache_free(net->cache, iocb);
+		if (unlikely(tx_total_len >= VHOST_NET_WEIGHT)) {
+			vhost_poll_queue(&vq->poll);
+			break;
+		}
+	}
+	spin_unlock_irqrestore(&vq->notify_lock, flags);
+}
+
+static struct kiocb *create_iocb(struct vhost_net *net,
+				 struct vhost_virtqueue *vq,
+				 unsigned head)
+{
+	struct kiocb *iocb = NULL;
+
+	if (!is_async_vq(vq))
+		return NULL;
+
+	iocb = kmem_cache_zalloc(net->cache, GFP_KERNEL);
+	if (!iocb)
+		return NULL;
+	iocb->private = vq;
+	iocb->ki_pos = head;
+	iocb->ki_dtor = handle_iocb;
+	if (vq == &net->dev.vqs[VHOST_NET_VQ_RX])
+		iocb->ki_user_data = vq->num;
+	iocb->ki_iovec = vq->hdr;
+	return iocb;
+}
+
 /* Expects to be always run from workqueue - which acts as
  * read-size critical section for our kind of RCU. */
 static void handle_tx(struct vhost_net *net)
 {
 	struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_TX];
+	struct kiocb *iocb = NULL;
 	unsigned head, out, in, s;
 	struct msghdr msg = {
 		.msg_name = NULL,
@@ -130,6 +314,8 @@ static void handle_tx(struct vhost_net *net)
 		tx_poll_stop(net);
 	vhost_hlen = vq->vhost_hlen;
 
+	handle_async_tx_events_notify(net, vq);
+
 	for (;;) {
 		head = vhost_get_desc(&net->dev, vq, vq->iov,
 				      ARRAY_SIZE(vq->iov),
@@ -138,10 +324,13 @@ static void handle_tx(struct vhost_net *net)
 		/* Nothing new?  Wait for eventfd to tell us they refilled. */
 		if (head == vq->num) {
 			wmem = atomic_read(&sock->sk->sk_wmem_alloc);
-			if (wmem >= sock->sk->sk_sndbuf * 3 / 4) {
-				tx_poll_start(net, sock);
-				set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
-				break;
+			if (!is_async_vq(vq)) {
+				if (wmem >= sock->sk->sk_sndbuf * 3 / 4) {
+					tx_poll_start(net, sock);
+					set_bit(SOCK_ASYNC_NOSPACE,
+						&sock->flags);
+					break;
+				}
 			}
 			if (unlikely(vhost_enable_notify(vq))) {
 				vhost_disable_notify(vq);
@@ -158,6 +347,13 @@ static void handle_tx(struct vhost_net *net)
 		s = move_iovec_hdr(vq->iov, vq->hdr, vhost_hlen, out);
 		msg.msg_iovlen = out;
 		len = iov_length(vq->iov, out);
+
+		if (is_async_vq(vq)) {
+			iocb = create_iocb(net, vq, head);
+			if (!iocb)
+				break;
+		}
+
 		/* Sanity check */
 		if (!len) {
 			vq_err(vq, "Unexpected header len for TX: "
@@ -166,12 +362,18 @@ static void handle_tx(struct vhost_net *net)
 			break;
 		}
 		/* TODO: Check specific error and bomb out unless ENOBUFS? */
-		err = sock->ops->sendmsg(NULL, sock, &msg, len);
+		err = sock->ops->sendmsg(iocb, sock, &msg, len);
 		if (unlikely(err < 0)) {
+			if (is_async_vq(vq))
+				kmem_cache_free(net->cache, iocb);
 			vhost_discard_desc(vq, 1);
 			tx_poll_start(net, sock);
 			break;
 		}
+
+		if (is_async_vq(vq))
+			continue;
+
 		if (err != len)
 			pr_err("Truncated TX packet: "
 			       " len %d != %zd\n", err, len);
@@ -183,6 +385,8 @@ static void handle_tx(struct vhost_net *net)
 		}
 	}
 
+	handle_async_tx_events_notify(net, vq);
+
 	mutex_unlock(&vq->mutex);
 	unuse_mm(net->dev.mm);
 }
@@ -205,7 +409,8 @@ static int vhost_head_len(struct vhost_virtqueue *vq, struct sock *sk)
 static void handle_rx(struct vhost_net *net)
 {
 	struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX];
-	unsigned in, log, s;
+	struct kiocb *iocb = NULL;
+	unsigned in, out, log, s;
 	struct vhost_log *vq_log;
 	struct msghdr msg = {
 		.msg_name = NULL,
@@ -225,25 +430,42 @@ static void handle_rx(struct vhost_net *net)
 	int err, headcount, datalen;
 	size_t vhost_hlen;
 	struct socket *sock = rcu_dereference(vq->private_data);
-	if (!sock || skb_queue_empty(&sock->sk->sk_receive_queue))
+	if (!sock || (skb_queue_empty(&sock->sk->sk_receive_queue) &&
+		      !is_async_vq(vq)))
 		return;
-
 	use_mm(net->dev.mm);
 	mutex_lock(&vq->mutex);
 	vhost_disable_notify(vq);
 	vhost_hlen = vq->vhost_hlen;
 
+	/* In async cases, when write log is enabled, in case the submitted
+	 * buffers did not get log info before the log enabling, so we'd
+	 * better recompute the log info when needed. We do this in
+	 * handle_async_rx_events_notify().
+	 */
+
 	vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ?
 		vq->log : NULL;
 
-	while ((datalen = vhost_head_len(vq, sock->sk))) {
-		headcount = vhost_get_desc_n(vq, vq->heads,
-					     datalen + vhost_hlen,
-					     &in, vq_log, &log);
+	handle_async_rx_events_notify(net, vq, sock);
+
+	while (is_async_vq(vq) ||
+		(datalen = vhost_head_len(vq, sock->sk)) != 0) {
+		if (is_async_vq(vq))
+			headcount =
+				vhost_get_desc(&net->dev, vq, vq->iov,
+						ARRAY_SIZE(vq->iov),
+						&out, &in,
+						vq->log, &log);
+		else
+			headcount = vhost_get_desc_n(vq, vq->heads,
+						     datalen + vhost_hlen,
+						     &in, vq_log, &log);
 		if (headcount < 0)
 			break;
 		/* OK, now we need to know about added descriptors. */
-		if (!headcount) {
+		if ((!headcount && !is_async_vq(vq)) ||
+			(headcount == vq->num && is_async_vq(vq))) {
 			if (unlikely(vhost_enable_notify(vq))) {
 				/* They have slipped one in as we were
 				 * doing that: check again. */
@@ -256,7 +478,12 @@ static void handle_rx(struct vhost_net *net)
 		}
 		/* We don't need to be notified again. */
 		/* Skip header. TODO: support TSO. */
+		if (is_async_vq(vq) && vhost_hlen == sizeof(hdr)) {
+			vq->hdr[0].iov_len = vhost_hlen;
+			goto nomove;
+		}
 		s = move_iovec_hdr(vq->iov, vq->hdr, vhost_hlen, in);
+nomove:
 		msg.msg_iovlen = in;
 		len = iov_length(vq->iov, in);
 		/* Sanity check */
@@ -266,13 +493,23 @@ static void handle_rx(struct vhost_net *net)
 			       iov_length(vq->hdr, s), vhost_hlen);
 			break;
 		}
-		err = sock->ops->recvmsg(NULL, sock, &msg,
+		if (is_async_vq(vq)) {
+			iocb = create_iocb(net, vq, headcount);
+			if (!iocb)
+				break;
+		}
+		err = sock->ops->recvmsg(iocb, sock, &msg,
 					 len, MSG_DONTWAIT | MSG_TRUNC);
 		/* TODO: Check specific error and bomb out unless EAGAIN? */
 		if (err < 0) {
+			if (is_async_vq(vq))
+				kmem_cache_free(net->cache, iocb);
 			vhost_discard_desc(vq, headcount);
 			break;
 		}
+		if (is_async_vq(vq))
+			continue;
+
 		if (err != datalen) {
 			pr_err("Discarded rx packet: "
 			       " len %d, expected %zd\n", err, datalen);
@@ -280,6 +517,9 @@ static void handle_rx(struct vhost_net *net)
 			continue;
 		}
 		len = err;
+		if (vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF))
+			hdr.num_buffers = headcount;
+
 		err = memcpy_toiovec(vq->hdr, (unsigned char *)&hdr,
 				     vhost_hlen);
 		if (err) {
@@ -287,18 +527,7 @@ static void handle_rx(struct vhost_net *net)
 			       vq->iov->iov_base, err);
 			break;
 		}
-		/* TODO: Should check and handle checksum. */
-		if (vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF)) {
-			struct iovec *iov = vhost_hlen ? vq->hdr : vq->iov;
-
-			if (memcpy_toiovecend(iov, (unsigned char *)&headcount,
-				      offsetof(typeof(hdr), num_buffers),
-				      sizeof(hdr.num_buffers))) {
-				vq_err(vq, "Failed num_buffers write");
-				vhost_discard_desc(vq, headcount);
-				break;
-			}
-		}
+
 		len += vhost_hlen;
 		vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
 					    headcount);
@@ -311,6 +540,8 @@ static void handle_rx(struct vhost_net *net)
 		}
 	}
 
+	handle_async_rx_events_notify(net, vq, sock);
+
 	mutex_unlock(&vq->mutex);
 	unuse_mm(net->dev.mm);
 }
@@ -364,6 +595,7 @@ static int vhost_net_open(struct inode *inode, struct file *f)
 	vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT);
 	vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN);
 	n->tx_poll_state = VHOST_NET_POLL_DISABLED;
+	n->cache = NULL;
 
 	f->private_data = n;
 
@@ -427,6 +659,21 @@ static void vhost_net_flush(struct vhost_net *n)
 	vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
 }
 
+static void vhost_async_cleanup(struct vhost_net *n)
+{
+	/* clean the notifier */
+	struct vhost_virtqueue *vq;
+	struct kiocb *iocb = NULL;
+	if (n->cache) {
+		vq = &n->dev.vqs[VHOST_NET_VQ_RX];
+		while ((iocb = notify_dequeue(vq)) != NULL)
+			kmem_cache_free(n->cache, iocb);
+		vq = &n->dev.vqs[VHOST_NET_VQ_TX];
+		while ((iocb = notify_dequeue(vq)) != NULL)
+			kmem_cache_free(n->cache, iocb);
+	}
+}
+
 static int vhost_net_release(struct inode *inode, struct file *f)
 {
 	struct vhost_net *n = f->private_data;
@@ -443,6 +690,7 @@ static int vhost_net_release(struct inode *inode, struct file *f)
 	/* We do an extra flush before freeing memory,
 	 * since jobs can re-queue themselves. */
 	vhost_net_flush(n);
+	vhost_async_cleanup(n);
 	kfree(n);
 	return 0;
 }
@@ -494,21 +742,58 @@ static struct socket *get_tap_socket(int fd)
 	return sock;
 }
 
-static struct socket *get_socket(int fd)
+static struct socket *get_mp_socket(int fd)
+{
+	struct file *file = fget(fd);
+	struct socket *sock;
+	if (!file)
+		return ERR_PTR(-EBADF);
+	sock = mp_get_socket(file);
+	if (IS_ERR(sock))
+		fput(file);
+	return sock;
+}
+
+static struct socket *get_socket(struct vhost_virtqueue *vq, int fd,
+				 enum vhost_vq_link_state *state)
 {
 	struct socket *sock;
 	/* special case to disable backend */
 	if (fd == -1)
 		return NULL;
+
+	*state = VHOST_VQ_LINK_SYNC;
+
 	sock = get_raw_socket(fd);
 	if (!IS_ERR(sock))
 		return sock;
 	sock = get_tap_socket(fd);
 	if (!IS_ERR(sock))
 		return sock;
+	/* If we dont' have notify_cache, then dont do mpassthru */
+	if (!notify_cache)
+		return ERR_PTR(-ENOTSOCK);
+	sock = get_mp_socket(fd);
+	if (!IS_ERR(sock)) {
+		*state = VHOST_VQ_LINK_ASYNC;
+		return sock;
+	}
 	return ERR_PTR(-ENOTSOCK);
 }
 
+static void vhost_init_link_state(struct vhost_net *n, int index)
+{
+	struct vhost_virtqueue *vq = n->vqs + index;
+
+	WARN_ON(!mutex_is_locked(&vq->mutex));
+	if (vq->link_state == VHOST_VQ_LINK_ASYNC) {
+		INIT_LIST_HEAD(&vq->notifier);
+		spin_lock_init(&vq->notify_lock);
+		if (!n->cache)
+			n->cache = notify_cache;
+	}
+}
+
 static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
 {
 	struct socket *sock, *oldsock;
@@ -532,12 +817,14 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
 		r = -EFAULT;
 		goto err_vq;
 	}
-	sock = get_socket(fd);
+	sock = get_socket(vq, fd, &vq->link_state);
 	if (IS_ERR(sock)) {
 		r = PTR_ERR(sock);
 		goto err_vq;
 	}
 
+	vhost_init_link_state(n, index);
+
 	/* start polling new socket */
 	oldsock = vq->private_data;
 	if (sock == oldsock)
@@ -687,6 +974,9 @@ int vhost_net_init(void)
 	r = misc_register(&vhost_net_misc);
 	if (r)
 		goto err_reg;
+	notify_cache = kmem_cache_create("vhost_kiocb",
+					sizeof(struct kiocb), 0,
+					SLAB_HWCACHE_ALIGN, NULL);
 	return 0;
 err_reg:
 	vhost_cleanup();
@@ -700,6 +990,8 @@ void vhost_net_exit(void)
 {
 	misc_deregister(&vhost_net_misc);
 	vhost_cleanup();
+	if (notify_cache)
+		kmem_cache_destroy(notify_cache);
 }
 module_exit(vhost_net_exit);
 
diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 118c8e0..66ff5c5 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -909,6 +909,85 @@ err:
 	return r;
 }
 
+unsigned __vhost_get_desc(struct vhost_dev *dev, struct vhost_virtqueue *vq,
+			   struct iovec iov[], unsigned int iov_size,
+			   unsigned int *out_num, unsigned int *in_num,
+			   struct vhost_log *log, unsigned int *log_num,
+			   unsigned int head)
+{
+	struct vring_desc desc;
+	unsigned int i, found = 0;
+	u16 last_avail_idx;
+	int ret;
+
+	/* When we start there are none of either input nor output. */
+	*out_num = *in_num = 0;
+	if (unlikely(log))
+		*log_num = 0;
+
+	i = head;
+	do {
+		unsigned iov_count = *in_num + *out_num;
+		if (i >= vq->num) {
+			vq_err(vq, "Desc index is %u > %u, head = %u",
+			       i, vq->num, head);
+			return vq->num;
+		}
+		if (++found > vq->num) {
+			vq_err(vq, "Loop detected: last one at %u "
+			       "vq size %u head %u\n",
+			       i, vq->num, head);
+			return vq->num;
+		}
+		ret = copy_from_user(&desc, vq->desc + i, sizeof desc);
+		if (ret) {
+			vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
+			       i, vq->desc + i);
+			return vq->num;
+		}
+		if (desc.flags & VRING_DESC_F_INDIRECT) {
+			ret = get_indirect(dev, vq, iov, iov_size,
+					   out_num, in_num,
+					   log, log_num, &desc);
+			if (ret < 0) {
+				vq_err(vq, "Failure detected "
+				       "in indirect descriptor at idx %d\n", i);
+				return vq->num;
+			}
+			continue;
+		}
+
+		ret = translate_desc(dev, desc.addr, desc.len, iov + iov_count,
+				     iov_size - iov_count);
+		if (ret < 0) {
+			vq_err(vq, "Translation failure %d descriptor idx %d\n",
+			       ret, i);
+			return vq->num;
+		}
+		if (desc.flags & VRING_DESC_F_WRITE) {
+			/* If this is an input descriptor,
+			 * increment that count. */
+			*in_num += ret;
+			if (unlikely(log)) {
+				log[*log_num].addr = desc.addr;
+				log[*log_num].len = desc.len;
+				++*log_num;
+			}
+		} else {
+			/* If it's an output descriptor, they're all supposed
+			 * to come before any input descriptors. */
+			if (*in_num) {
+				vq_err(vq, "Descriptor has out after in: "
+				       "idx %d\n", i);
+				return vq->num;
+			}
+			*out_num += ret;
+		}
+	} while ((i = next_desc(&desc)) != -1);
+
+	return head;
+}
+
 /* This looks in the virtqueue and for the first available buffer, and converts
  * it to an iovec for convenient access.  Since descriptors consist of some
  * number of output then some number of input descriptors, it's actually two
diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index 08d740a..54c6d0b 100644
--- a/drivers/vhost/vhost.h
+++ b/drivers/vhost/vhost.h
@@ -43,6 +43,11 @@ struct vhost_log {
 	u64 len;
 };
 
+enum vhost_vq_link_state {
+	VHOST_VQ_LINK_SYNC = 0,
+	VHOST_VQ_LINK_ASYNC = 1,
+};
+
 /* The virtqueue structure describes a queue attached to a device. */
 struct vhost_virtqueue {
 	struct vhost_dev *dev;
@@ -98,6 +103,10 @@ struct vhost_virtqueue {
 	/* Log write descriptors */
 	void __user *log_base;
 	struct vhost_log log[VHOST_NET_MAX_SG];
+	/* Differiate async socket for 0-copy from normal */
+	enum vhost_vq_link_state link_state;
+	struct list_head notifier;
+	spinlock_t notify_lock;
 };
 
 struct vhost_dev {
@@ -125,6 +134,11 @@ int vhost_log_access_ok(struct vhost_dev *);
 int vhost_get_desc_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
 		     int datalen, unsigned int *iovcount, struct vhost_log *log,
 		     unsigned int *log_num);
+unsigned __vhost_get_desc(struct vhost_dev *, struct vhost_virtqueue *,
+			struct iovec iov[], unsigned int iov_count,
+			unsigned int *out_num, unsigned int *in_num,
+			struct vhost_log *log, unsigned int *log_num,
+			unsigned int head);
 unsigned vhost_get_desc(struct vhost_dev *, struct vhost_virtqueue *,
 			   struct iovec iov[], unsigned int iov_count,
 			   unsigned int *out_num, unsigned int *in_num,
@@ -165,6 +179,7 @@ enum {
 static inline int vhost_has_feature(struct vhost_dev *dev, int bit)
 {
 	unsigned acked_features = rcu_dereference(dev->acked_features);
+	acked_features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
 	return acked_features & (1 << bit);
 }
 
-- 
1.5.4.4

[RFC PATCH v10 14/16] Provides multiple submits and asynchronous notifications.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

The vhost-net backend now only supports synchronous send/recv
operations. The patch provides multiple submits and asynchronous
notifications. This is needed for zero-copy case.

Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
---
 drivers/vhost/net.c   |  348 +++++++++++++++++++++++++++++++++++++++++++++----
 drivers/vhost/vhost.c |   79 +++++++++++
 drivers/vhost/vhost.h |   15 ++
 3 files changed, 414 insertions(+), 28 deletions(-)

diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index b38abc6..c4bc815 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -24,6 +24,8 @@
 #include <linux/if_arp.h>
 #include <linux/if_tun.h>
 #include <linux/if_macvlan.h>
+#include <linux/mpassthru.h>
+#include <linux/aio.h>
 
 #include <net/sock.h>
 
@@ -39,6 +41,8 @@ enum {
 	VHOST_NET_VQ_MAX = 2,
 };
 
+static struct kmem_cache *notify_cache;
+
 enum vhost_net_poll_state {
 	VHOST_NET_POLL_DISABLED = 0,
 	VHOST_NET_POLL_STARTED = 1,
@@ -49,6 +53,7 @@ struct vhost_net {
 	struct vhost_dev dev;
 	struct vhost_virtqueue vqs[VHOST_NET_VQ_MAX];
 	struct vhost_poll poll[VHOST_NET_VQ_MAX];
+	struct kmem_cache       *cache;
 	/* Tells us whether we are polling a socket for TX.
 	 * We only do this when socket buffer fills up.
 	 * Protected by tx vq lock. */
@@ -93,11 +98,190 @@ static void tx_poll_start(struct vhost_net *net, struct socket *sock)
 	net->tx_poll_state = VHOST_NET_POLL_STARTED;
 }
 
+struct kiocb *notify_dequeue(struct vhost_virtqueue *vq)
+{
+	struct kiocb *iocb = NULL;
+	unsigned long flags;
+
+	spin_lock_irqsave(&vq->notify_lock, flags);
+	if (!list_empty(&vq->notifier)) {
+		iocb = list_first_entry(&vq->notifier,
+				struct kiocb, ki_list);
+		list_del(&iocb->ki_list);
+	}
+	spin_unlock_irqrestore(&vq->notify_lock, flags);
+	return iocb;
+}
+
+static void handle_iocb(struct kiocb *iocb)
+{
+	struct vhost_virtqueue *vq = iocb->private;
+	unsigned long flags;
+
+	spin_lock_irqsave(&vq->notify_lock, flags);
+	list_add_tail(&iocb->ki_list, &vq->notifier);
+	spin_unlock_irqrestore(&vq->notify_lock, flags);
+}
+
+static int is_async_vq(struct vhost_virtqueue *vq)
+{
+	return (vq->link_state == VHOST_VQ_LINK_ASYNC);
+}
+
+static void handle_async_rx_events_notify(struct vhost_net *net,
+					  struct vhost_virtqueue *vq,
+					  struct socket *sock)
+{
+	struct kiocb *iocb = NULL;
+	struct vhost_log *vq_log = NULL;
+	int rx_total_len = 0;
+	unsigned int head, log, in, out;
+	int size;
+	int count;
+
+	struct virtio_net_hdr_mrg_rxbuf hdr = {
+		.hdr.flags = 0,
+		.hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE
+	};
+
+	if (!is_async_vq(vq))
+		return;
+
+	if (sock->sk->sk_data_ready)
+		sock->sk->sk_data_ready(sock->sk, 0);
+
+	vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ?
+		vq->log : NULL;
+
+	while ((iocb = notify_dequeue(vq)) != NULL) {
+		if (!iocb->ki_left) {
+			vhost_add_used_and_signal(&net->dev, vq,
+					iocb->ki_pos, iocb->ki_nbytes);
+			size = iocb->ki_nbytes;
+			head = iocb->ki_pos;
+			rx_total_len += iocb->ki_nbytes;
+
+			if (iocb->ki_dtor)
+				iocb->ki_dtor(iocb);
+			kmem_cache_free(net->cache, iocb);
+
+			/* when log is enabled, recomputing the log is needed,
+			 * since these buffers are in async queue, may not get
+			 * the log info before.
+			 */
+			if (unlikely(vq_log)) {
+				if (!log)
+					__vhost_get_desc(&net->dev, vq, vq->iov,
+							ARRAY_SIZE(vq->iov),
+							&out, &in, vq_log,
+							&log, head);
+				vhost_log_write(vq, vq_log, log, size);
+			}
+			if (unlikely(rx_total_len >= VHOST_NET_WEIGHT)) {
+				vhost_poll_queue(&vq->poll);
+				break;
+			}
+		} else {
+			int i = 0;
+			int count = iocb->ki_left;
+			int hc = count;
+			while (count--) {
+				if (iocb) {
+					vq->heads[i].id = iocb->ki_pos;
+					vq->heads[i].len = iocb->ki_nbytes;
+					size = iocb->ki_nbytes;
+					head = iocb->ki_pos;
+					rx_total_len += iocb->ki_nbytes;
+
+					if (iocb->ki_dtor)
+						iocb->ki_dtor(iocb);
+					kmem_cache_free(net->cache, iocb);
+
+					if (unlikely(vq_log)) {
+						if (!log)
+							__vhost_get_desc(
+							&net->dev, vq, vq->iov,
+							ARRAY_SIZE(vq->iov),
+							&out, &in, vq_log,
+							&log, head);
+						vhost_log_write(
+							vq, vq_log, log, size);
+					}
+				} else
+					break;
+
+				i++;
+				iocb == NULL;
+				if (count)
+					iocb = notify_dequeue(vq);
+			}
+			vhost_add_used_and_signal_n(
+					&net->dev, vq, vq->heads, hc);
+		}
+	}
+}
+
+static void handle_async_tx_events_notify(struct vhost_net *net,
+					  struct vhost_virtqueue *vq)
+{
+	struct kiocb *iocb = NULL;
+	struct list_head *entry, *tmp;
+	unsigned long flags;
+	int tx_total_len = 0;
+
+	if (!is_async_vq(vq))
+		return;
+
+	spin_lock_irqsave(&vq->notify_lock, flags);
+	list_for_each_safe(entry, tmp, &vq->notifier) {
+		iocb = list_entry(entry,
+				  struct kiocb, ki_list);
+		if (!iocb->ki_flags)
+			continue;
+		list_del(&iocb->ki_list);
+		vhost_add_used_and_signal(&net->dev, vq,
+				iocb->ki_pos, 0);
+		tx_total_len += iocb->ki_nbytes;
+
+		if (iocb->ki_dtor)
+			iocb->ki_dtor(iocb);
+
+		kmem_cache_free(net->cache, iocb);
+		if (unlikely(tx_total_len >= VHOST_NET_WEIGHT)) {
+			vhost_poll_queue(&vq->poll);
+			break;
+		}
+	}
+	spin_unlock_irqrestore(&vq->notify_lock, flags);
+}
+
+static struct kiocb *create_iocb(struct vhost_net *net,
+				 struct vhost_virtqueue *vq,
+				 unsigned head)
+{
+	struct kiocb *iocb = NULL;
+
+	if (!is_async_vq(vq))
+		return NULL;
+
+	iocb = kmem_cache_zalloc(net->cache, GFP_KERNEL);
+	if (!iocb)
+		return NULL;
+	iocb->private = vq;
+	iocb->ki_pos = head;
+	iocb->ki_dtor = handle_iocb;
+	if (vq == &net->dev.vqs[VHOST_NET_VQ_RX])
+		iocb->ki_user_data = vq->num;
+	iocb->ki_iovec = vq->hdr;
+	return iocb;
+}
+
 /* Expects to be always run from workqueue - which acts as
  * read-size critical section for our kind of RCU. */
 static void handle_tx(struct vhost_net *net)
 {
 	struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_TX];
+	struct kiocb *iocb = NULL;
 	unsigned head, out, in, s;
 	struct msghdr msg = {
 		.msg_name = NULL,
@@ -130,6 +314,8 @@ static void handle_tx(struct vhost_net *net)
 		tx_poll_stop(net);
 	vhost_hlen = vq->vhost_hlen;
 
+	handle_async_tx_events_notify(net, vq);
+
 	for (;;) {
 		head = vhost_get_desc(&net->dev, vq, vq->iov,
 				      ARRAY_SIZE(vq->iov),
@@ -138,10 +324,13 @@ static void handle_tx(struct vhost_net *net)
 		/* Nothing new?  Wait for eventfd to tell us they refilled. */
 		if (head == vq->num) {
 			wmem = atomic_read(&sock->sk->sk_wmem_alloc);
-			if (wmem >= sock->sk->sk_sndbuf * 3 / 4) {
-				tx_poll_start(net, sock);
-				set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
-				break;
+			if (!is_async_vq(vq)) {
+				if (wmem >= sock->sk->sk_sndbuf * 3 / 4) {
+					tx_poll_start(net, sock);
+					set_bit(SOCK_ASYNC_NOSPACE,
+						&sock->flags);
+					break;
+				}
 			}
 			if (unlikely(vhost_enable_notify(vq))) {
 				vhost_disable_notify(vq);
@@ -158,6 +347,13 @@ static void handle_tx(struct vhost_net *net)
 		s = move_iovec_hdr(vq->iov, vq->hdr, vhost_hlen, out);
 		msg.msg_iovlen = out;
 		len = iov_length(vq->iov, out);
+
+		if (is_async_vq(vq)) {
+			iocb = create_iocb(net, vq, head);
+			if (!iocb)
+				break;
+		}
+
 		/* Sanity check */
 		if (!len) {
 			vq_err(vq, "Unexpected header len for TX: "
@@ -166,12 +362,18 @@ static void handle_tx(struct vhost_net *net)
 			break;
 		}
 		/* TODO: Check specific error and bomb out unless ENOBUFS? */
-		err = sock->ops->sendmsg(NULL, sock, &msg, len);
+		err = sock->ops->sendmsg(iocb, sock, &msg, len);
 		if (unlikely(err < 0)) {
+			if (is_async_vq(vq))
+				kmem_cache_free(net->cache, iocb);
 			vhost_discard_desc(vq, 1);
 			tx_poll_start(net, sock);
 			break;
 		}
+
+		if (is_async_vq(vq))
+			continue;
+
 		if (err != len)
 			pr_err("Truncated TX packet: "
 			       " len %d != %zd\n", err, len);
@@ -183,6 +385,8 @@ static void handle_tx(struct vhost_net *net)
 		}
 	}
 
+	handle_async_tx_events_notify(net, vq);
+
 	mutex_unlock(&vq->mutex);
 	unuse_mm(net->dev.mm);
 }
@@ -205,7 +409,8 @@ static int vhost_head_len(struct vhost_virtqueue *vq, struct sock *sk)
 static void handle_rx(struct vhost_net *net)
 {
 	struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX];
-	unsigned in, log, s;
+	struct kiocb *iocb = NULL;
+	unsigned in, out, log, s;
 	struct vhost_log *vq_log;
 	struct msghdr msg = {
 		.msg_name = NULL,
@@ -225,25 +430,42 @@ static void handle_rx(struct vhost_net *net)
 	int err, headcount, datalen;
 	size_t vhost_hlen;
 	struct socket *sock = rcu_dereference(vq->private_data);
-	if (!sock || skb_queue_empty(&sock->sk->sk_receive_queue))
+	if (!sock || (skb_queue_empty(&sock->sk->sk_receive_queue) &&
+		      !is_async_vq(vq)))
 		return;
-
 	use_mm(net->dev.mm);
 	mutex_lock(&vq->mutex);
 	vhost_disable_notify(vq);
 	vhost_hlen = vq->vhost_hlen;
 
+	/* In async cases, when write log is enabled, in case the submitted
+	 * buffers did not get log info before the log enabling, so we'd
+	 * better recompute the log info when needed. We do this in
+	 * handle_async_rx_events_notify().
+	 */
+
 	vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ?
 		vq->log : NULL;
 
-	while ((datalen = vhost_head_len(vq, sock->sk))) {
-		headcount = vhost_get_desc_n(vq, vq->heads,
-					     datalen + vhost_hlen,
-					     &in, vq_log, &log);
+	handle_async_rx_events_notify(net, vq, sock);
+
+	while (is_async_vq(vq) ||
+		(datalen = vhost_head_len(vq, sock->sk)) != 0) {
+		if (is_async_vq(vq))
+			headcount =
+				vhost_get_desc(&net->dev, vq, vq->iov,
+						ARRAY_SIZE(vq->iov),
+						&out, &in,
+						vq->log, &log);
+		else
+			headcount = vhost_get_desc_n(vq, vq->heads,
+						     datalen + vhost_hlen,
+						     &in, vq_log, &log);
 		if (headcount < 0)
 			break;
 		/* OK, now we need to know about added descriptors. */
-		if (!headcount) {
+		if ((!headcount && !is_async_vq(vq)) ||
+			(headcount == vq->num && is_async_vq(vq))) {
 			if (unlikely(vhost_enable_notify(vq))) {
 				/* They have slipped one in as we were
 				 * doing that: check again. */
@@ -256,7 +478,12 @@ static void handle_rx(struct vhost_net *net)
 		}
 		/* We don't need to be notified again. */
 		/* Skip header. TODO: support TSO. */
+		if (is_async_vq(vq) && vhost_hlen == sizeof(hdr)) {
+			vq->hdr[0].iov_len = vhost_hlen;
+			goto nomove;
+		}
 		s = move_iovec_hdr(vq->iov, vq->hdr, vhost_hlen, in);
+nomove:
 		msg.msg_iovlen = in;
 		len = iov_length(vq->iov, in);
 		/* Sanity check */
@@ -266,13 +493,23 @@ static void handle_rx(struct vhost_net *net)
 			       iov_length(vq->hdr, s), vhost_hlen);
 			break;
 		}
-		err = sock->ops->recvmsg(NULL, sock, &msg,
+		if (is_async_vq(vq)) {
+			iocb = create_iocb(net, vq, headcount);
+			if (!iocb)
+				break;
+		}
+		err = sock->ops->recvmsg(iocb, sock, &msg,
 					 len, MSG_DONTWAIT | MSG_TRUNC);
 		/* TODO: Check specific error and bomb out unless EAGAIN? */
 		if (err < 0) {
+			if (is_async_vq(vq))
+				kmem_cache_free(net->cache, iocb);
 			vhost_discard_desc(vq, headcount);
 			break;
 		}
+		if (is_async_vq(vq))
+			continue;
+
 		if (err != datalen) {
 			pr_err("Discarded rx packet: "
 			       " len %d, expected %zd\n", err, datalen);
@@ -280,6 +517,9 @@ static void handle_rx(struct vhost_net *net)
 			continue;
 		}
 		len = err;
+		if (vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF))
+			hdr.num_buffers = headcount;
+
 		err = memcpy_toiovec(vq->hdr, (unsigned char *)&hdr,
 				     vhost_hlen);
 		if (err) {
@@ -287,18 +527,7 @@ static void handle_rx(struct vhost_net *net)
 			       vq->iov->iov_base, err);
 			break;
 		}
-		/* TODO: Should check and handle checksum. */
-		if (vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF)) {
-			struct iovec *iov = vhost_hlen ? vq->hdr : vq->iov;
-
-			if (memcpy_toiovecend(iov, (unsigned char *)&headcount,
-				      offsetof(typeof(hdr), num_buffers),
-				      sizeof(hdr.num_buffers))) {
-				vq_err(vq, "Failed num_buffers write");
-				vhost_discard_desc(vq, headcount);
-				break;
-			}
-		}
+
 		len += vhost_hlen;
 		vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
 					    headcount);
@@ -311,6 +540,8 @@ static void handle_rx(struct vhost_net *net)
 		}
 	}
 
+	handle_async_rx_events_notify(net, vq, sock);
+
 	mutex_unlock(&vq->mutex);
 	unuse_mm(net->dev.mm);
 }
@@ -364,6 +595,7 @@ static int vhost_net_open(struct inode *inode, struct file *f)
 	vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT);
 	vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN);
 	n->tx_poll_state = VHOST_NET_POLL_DISABLED;
+	n->cache = NULL;
 
 	f->private_data = n;
 
@@ -427,6 +659,21 @@ static void vhost_net_flush(struct vhost_net *n)
 	vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
 }
 
+static void vhost_async_cleanup(struct vhost_net *n)
+{
+	/* clean the notifier */
+	struct vhost_virtqueue *vq;
+	struct kiocb *iocb = NULL;
+	if (n->cache) {
+		vq = &n->dev.vqs[VHOST_NET_VQ_RX];
+		while ((iocb = notify_dequeue(vq)) != NULL)
+			kmem_cache_free(n->cache, iocb);
+		vq = &n->dev.vqs[VHOST_NET_VQ_TX];
+		while ((iocb = notify_dequeue(vq)) != NULL)
+			kmem_cache_free(n->cache, iocb);
+	}
+}
+
 static int vhost_net_release(struct inode *inode, struct file *f)
 {
 	struct vhost_net *n = f->private_data;
@@ -443,6 +690,7 @@ static int vhost_net_release(struct inode *inode, struct file *f)
 	/* We do an extra flush before freeing memory,
 	 * since jobs can re-queue themselves. */
 	vhost_net_flush(n);
+	vhost_async_cleanup(n);
 	kfree(n);
 	return 0;
 }
@@ -494,21 +742,58 @@ static struct socket *get_tap_socket(int fd)
 	return sock;
 }
 
-static struct socket *get_socket(int fd)
+static struct socket *get_mp_socket(int fd)
+{
+	struct file *file = fget(fd);
+	struct socket *sock;
+	if (!file)
+		return ERR_PTR(-EBADF);
+	sock = mp_get_socket(file);
+	if (IS_ERR(sock))
+		fput(file);
+	return sock;
+}
+
+static struct socket *get_socket(struct vhost_virtqueue *vq, int fd,
+				 enum vhost_vq_link_state *state)
 {
 	struct socket *sock;
 	/* special case to disable backend */
 	if (fd == -1)
 		return NULL;
+
+	*state = VHOST_VQ_LINK_SYNC;
+
 	sock = get_raw_socket(fd);
 	if (!IS_ERR(sock))
 		return sock;
 	sock = get_tap_socket(fd);
 	if (!IS_ERR(sock))
 		return sock;
+	/* If we dont' have notify_cache, then dont do mpassthru */
+	if (!notify_cache)
+		return ERR_PTR(-ENOTSOCK);
+	sock = get_mp_socket(fd);
+	if (!IS_ERR(sock)) {
+		*state = VHOST_VQ_LINK_ASYNC;
+		return sock;
+	}
 	return ERR_PTR(-ENOTSOCK);
 }
 
+static void vhost_init_link_state(struct vhost_net *n, int index)
+{
+	struct vhost_virtqueue *vq = n->vqs + index;
+
+	WARN_ON(!mutex_is_locked(&vq->mutex));
+	if (vq->link_state == VHOST_VQ_LINK_ASYNC) {
+		INIT_LIST_HEAD(&vq->notifier);
+		spin_lock_init(&vq->notify_lock);
+		if (!n->cache)
+			n->cache = notify_cache;
+	}
+}
+
 static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
 {
 	struct socket *sock, *oldsock;
@@ -532,12 +817,14 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
 		r = -EFAULT;
 		goto err_vq;
 	}
-	sock = get_socket(fd);
+	sock = get_socket(vq, fd, &vq->link_state);
 	if (IS_ERR(sock)) {
 		r = PTR_ERR(sock);
 		goto err_vq;
 	}
 
+	vhost_init_link_state(n, index);
+
 	/* start polling new socket */
 	oldsock = vq->private_data;
 	if (sock == oldsock)
@@ -687,6 +974,9 @@ int vhost_net_init(void)
 	r = misc_register(&vhost_net_misc);
 	if (r)
 		goto err_reg;
+	notify_cache = kmem_cache_create("vhost_kiocb",
+					sizeof(struct kiocb), 0,
+					SLAB_HWCACHE_ALIGN, NULL);
 	return 0;
 err_reg:
 	vhost_cleanup();
@@ -700,6 +990,8 @@ void vhost_net_exit(void)
 {
 	misc_deregister(&vhost_net_misc);
 	vhost_cleanup();
+	if (notify_cache)
+		kmem_cache_destroy(notify_cache);
 }
 module_exit(vhost_net_exit);
 
diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 118c8e0..66ff5c5 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -909,6 +909,85 @@ err:
 	return r;
 }
 
+unsigned __vhost_get_desc(struct vhost_dev *dev, struct vhost_virtqueue *vq,
+			   struct iovec iov[], unsigned int iov_size,
+			   unsigned int *out_num, unsigned int *in_num,
+			   struct vhost_log *log, unsigned int *log_num,
+			   unsigned int head)
+{
+	struct vring_desc desc;
+	unsigned int i, found = 0;
+	u16 last_avail_idx;
+	int ret;
+
+	/* When we start there are none of either input nor output. */
+	*out_num = *in_num = 0;
+	if (unlikely(log))
+		*log_num = 0;
+
+	i = head;
+	do {
+		unsigned iov_count = *in_num + *out_num;
+		if (i >= vq->num) {
+			vq_err(vq, "Desc index is %u > %u, head = %u",
+			       i, vq->num, head);
+			return vq->num;
+		}
+		if (++found > vq->num) {
+			vq_err(vq, "Loop detected: last one at %u "
+			       "vq size %u head %u\n",
+			       i, vq->num, head);
+			return vq->num;
+		}
+		ret = copy_from_user(&desc, vq->desc + i, sizeof desc);
+		if (ret) {
+			vq_err(vq, "Failed to get descriptor: idx %d addr %p\n",
+			       i, vq->desc + i);
+			return vq->num;
+		}
+		if (desc.flags & VRING_DESC_F_INDIRECT) {
+			ret = get_indirect(dev, vq, iov, iov_size,
+					   out_num, in_num,
+					   log, log_num, &desc);
+			if (ret < 0) {
+				vq_err(vq, "Failure detected "
+				       "in indirect descriptor at idx %d\n", i);
+				return vq->num;
+			}
+			continue;
+		}
+
+		ret = translate_desc(dev, desc.addr, desc.len, iov + iov_count,
+				     iov_size - iov_count);
+		if (ret < 0) {
+			vq_err(vq, "Translation failure %d descriptor idx %d\n",
+			       ret, i);
+			return vq->num;
+		}
+		if (desc.flags & VRING_DESC_F_WRITE) {
+			/* If this is an input descriptor,
+			 * increment that count. */
+			*in_num += ret;
+			if (unlikely(log)) {
+				log[*log_num].addr = desc.addr;
+				log[*log_num].len = desc.len;
+				++*log_num;
+			}
+		} else {
+			/* If it's an output descriptor, they're all supposed
+			 * to come before any input descriptors. */
+			if (*in_num) {
+				vq_err(vq, "Descriptor has out after in: "
+				       "idx %d\n", i);
+				return vq->num;
+			}
+			*out_num += ret;
+		}
+	} while ((i = next_desc(&desc)) != -1);
+
+	return head;
+}
+
 /* This looks in the virtqueue and for the first available buffer, and converts
  * it to an iovec for convenient access.  Since descriptors consist of some
  * number of output then some number of input descriptors, it's actually two
diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index 08d740a..54c6d0b 100644
--- a/drivers/vhost/vhost.h
+++ b/drivers/vhost/vhost.h
@@ -43,6 +43,11 @@ struct vhost_log {
 	u64 len;
 };
 
+enum vhost_vq_link_state {
+	VHOST_VQ_LINK_SYNC = 0,
+	VHOST_VQ_LINK_ASYNC = 1,
+};
+
 /* The virtqueue structure describes a queue attached to a device. */
 struct vhost_virtqueue {
 	struct vhost_dev *dev;
@@ -98,6 +103,10 @@ struct vhost_virtqueue {
 	/* Log write descriptors */
 	void __user *log_base;
 	struct vhost_log log[VHOST_NET_MAX_SG];
+	/* Differiate async socket for 0-copy from normal */
+	enum vhost_vq_link_state link_state;
+	struct list_head notifier;
+	spinlock_t notify_lock;
 };
 
 struct vhost_dev {
@@ -125,6 +134,11 @@ int vhost_log_access_ok(struct vhost_dev *);
 int vhost_get_desc_n(struct vhost_virtqueue *, struct vring_used_elem *heads,
 		     int datalen, unsigned int *iovcount, struct vhost_log *log,
 		     unsigned int *log_num);
+unsigned __vhost_get_desc(struct vhost_dev *, struct vhost_virtqueue *,
+			struct iovec iov[], unsigned int iov_count,
+			unsigned int *out_num, unsigned int *in_num,
+			struct vhost_log *log, unsigned int *log_num,
+			unsigned int head);
 unsigned vhost_get_desc(struct vhost_dev *, struct vhost_virtqueue *,
 			   struct iovec iov[], unsigned int iov_count,
 			   unsigned int *out_num, unsigned int *in_num,
@@ -165,6 +179,7 @@ enum {
 static inline int vhost_has_feature(struct vhost_dev *dev, int bit)
 {
 	unsigned acked_features = rcu_dereference(dev->acked_features);
+	acked_features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
 	return acked_features & (1 << bit);
 }
 
-- 
1.5.4.4

[RFC PATCH v10 15/16] An example how to modifiy NIC driver to use napi_gro_frags() interface

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

This example is made on ixgbe driver.
It provides API is_rx_buffer_mapped_as_page() to indicate
if the driver use napi_gro_frags() interface or not.
The example allocates 2 pages for DMA for one ring descriptor
using netdev_alloc_page(). When packets is coming, using
napi_gro_frags() to allocate skb and to receive the packets.

---
 drivers/net/ixgbe/ixgbe.h      |    3 +
 drivers/net/ixgbe/ixgbe_main.c |  151 ++++++++++++++++++++++++++++++++--------
 2 files changed, 125 insertions(+), 29 deletions(-)

diff --git a/drivers/net/ixgbe/ixgbe.h b/drivers/net/ixgbe/ixgbe.h
index 79c35ae..fceffc5 100644
--- a/drivers/net/ixgbe/ixgbe.h
+++ b/drivers/net/ixgbe/ixgbe.h
@@ -131,6 +131,9 @@ struct ixgbe_rx_buffer {
 	struct page *page;
 	dma_addr_t page_dma;
 	unsigned int page_offset;
+	u16 mapped_as_page;
+	struct page *page_skb;
+	unsigned int page_skb_offset;
 };
 
 struct ixgbe_queue_stats {
diff --git a/drivers/net/ixgbe/ixgbe_main.c b/drivers/net/ixgbe/ixgbe_main.c
index 6c00ee4..905d6d2 100644
--- a/drivers/net/ixgbe/ixgbe_main.c
+++ b/drivers/net/ixgbe/ixgbe_main.c
@@ -688,6 +688,12 @@ static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
 	IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
 }
 
+static bool is_rx_buffer_mapped_as_page(struct ixgbe_rx_buffer *bi,
+					struct net_device *dev)
+{
+	return true;
+}
+
 /**
  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
  * @adapter: address of board private structure
@@ -704,13 +710,17 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 	i = rx_ring->next_to_use;
 	bi = &rx_ring->rx_buffer_info[i];
 
+
 	while (cleaned_count--) {
 		rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
 
+		bi->mapped_as_page =
+			is_rx_buffer_mapped_as_page(bi, adapter->netdev);
+
 		if (!bi->page_dma &&
 		    (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
 			if (!bi->page) {
-				bi->page = alloc_page(GFP_ATOMIC);
+				bi->page = netdev_alloc_page(adapter->netdev);
 				if (!bi->page) {
 					adapter->alloc_rx_page_failed++;
 					goto no_buffers;
@@ -727,7 +737,7 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 			                            PCI_DMA_FROMDEVICE);
 		}
 
-		if (!bi->skb) {
+		if (!bi->mapped_as_page && !bi->skb) {
 			struct sk_buff *skb;
 			/* netdev_alloc_skb reserves 32 bytes up front!! */
 			uint bufsz = rx_ring->rx_buf_len + SMP_CACHE_BYTES;
@@ -747,6 +757,19 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 			                         rx_ring->rx_buf_len,
 			                         PCI_DMA_FROMDEVICE);
 		}
+
+		if (bi->mapped_as_page && !bi->page_skb) {
+			bi->page_skb = netdev_alloc_page(adapter->netdev);
+			if (!bi->page_skb) {
+				adapter->alloc_rx_page_failed++;
+				goto no_buffers;
+			}
+			bi->page_skb_offset = 0;
+			bi->dma = pci_map_page(pdev, bi->page_skb,
+					bi->page_skb_offset,
+					(PAGE_SIZE / 2),
+					PCI_DMA_FROMDEVICE);
+		}
 		/* Refresh the desc even if buffer_addrs didn't change because
 		 * each write-back erases this info. */
 		if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
@@ -823,6 +846,13 @@ struct ixgbe_rsc_cb {
 	dma_addr_t dma;
 };
 
+static bool is_no_buffer(struct ixgbe_rx_buffer *rx_buffer_info)
+{
+	return (!rx_buffer_info->skb ||
+		!rx_buffer_info->page_skb) &&
+		!rx_buffer_info->page;
+}
+
 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
 
 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
@@ -832,6 +862,7 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 	struct ixgbe_adapter *adapter = q_vector->adapter;
 	struct net_device *netdev = adapter->netdev;
 	struct pci_dev *pdev = adapter->pdev;
+	struct napi_struct *napi = &q_vector->napi;
 	union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
 	struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
 	struct sk_buff *skb;
@@ -868,29 +899,71 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 			len = le16_to_cpu(rx_desc->wb.upper.length);
 		}
 
+		if (is_no_buffer(rx_buffer_info))
+			break;
+
 		cleaned = true;
-		skb = rx_buffer_info->skb;
-		prefetch(skb->data);
-		rx_buffer_info->skb = NULL;
 
-		if (rx_buffer_info->dma) {
-			if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
-			    (!(staterr & IXGBE_RXD_STAT_EOP)) &&
-				 (!(skb->prev)))
-				/*
-				 * When HWRSC is enabled, delay unmapping
-				 * of the first packet. It carries the
-				 * header information, HW may still
-				 * access the header after the writeback.
-				 * Only unmap it when EOP is reached
-				 */
-				IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
-			else
-				pci_unmap_single(pdev, rx_buffer_info->dma,
-				                 rx_ring->rx_buf_len,
-				                 PCI_DMA_FROMDEVICE);
-			rx_buffer_info->dma = 0;
-			skb_put(skb, len);
+		if (!rx_buffer_info->mapped_as_page) {
+			skb = rx_buffer_info->skb;
+			prefetch(skb->data);
+			rx_buffer_info->skb = NULL;
+
+			if (rx_buffer_info->dma) {
+				if ((adapter->flags2 &
+					IXGBE_FLAG2_RSC_ENABLED) &&
+					(!(staterr & IXGBE_RXD_STAT_EOP)) &&
+					(!(skb->prev)))
+					/*
+					 * When HWRSC is enabled, delay unmapping
+					 * of the first packet. It carries the
+					 * header information, HW may still
+					 * access the header after the writeback.
+					 * Only unmap it when EOP is reached
+					 */
+					IXGBE_RSC_CB(skb)->dma =
+							rx_buffer_info->dma;
+				else
+					pci_unmap_single(pdev,
+							rx_buffer_info->dma,
+							rx_ring->rx_buf_len,
+							PCI_DMA_FROMDEVICE);
+				rx_buffer_info->dma = 0;
+				skb_put(skb, len);
+			}
+		} else {
+			skb = napi_get_frags(napi);
+			prefetch(rx_buffer_info->page_skb_offset);
+			rx_buffer_info->skb = NULL;
+			if (rx_buffer_info->dma) {
+				if ((adapter->flags2 &
+					IXGBE_FLAG2_RSC_ENABLED) &&
+					(!(staterr & IXGBE_RXD_STAT_EOP)) &&
+					(!(skb->prev)))
+					/*
+					 * When HWRSC is enabled, delay unmapping
+					 * of the first packet. It carries the
+					 * header information, HW may still
+					 * access the header after the writeback.
+					 * Only unmap it when EOP is reached
+					 */
+					IXGBE_RSC_CB(skb)->dma =
+						rx_buffer_info->dma;
+				else
+					pci_unmap_page(pdev, rx_buffer_info->dma,
+							PAGE_SIZE / 2,
+							PCI_DMA_FROMDEVICE);
+				rx_buffer_info->dma = 0;
+				skb_fill_page_desc(skb,
+						skb_shinfo(skb)->nr_frags,
+						rx_buffer_info->page_skb,
+						rx_buffer_info->page_skb_offset,
+						len);
+				rx_buffer_info->page_skb = NULL;
+				skb->len += len;
+				skb->data_len += len;
+				skb->truesize += len;
+			}
 		}
 
 		if (upper_len) {
@@ -956,6 +1029,11 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 				rx_buffer_info->dma = next_buffer->dma;
 				next_buffer->skb = skb;
 				next_buffer->dma = 0;
+				if (rx_buffer_info->mapped_as_page) {
+					rx_buffer_info->page_skb =
+							next_buffer->page_skb;
+					next_buffer->page_skb = NULL;
+				}
 			} else {
 				skb->next = next_buffer->skb;
 				skb->next->prev = skb;
@@ -975,7 +1053,8 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 		total_rx_bytes += skb->len;
 		total_rx_packets++;
 
-		skb->protocol = eth_type_trans(skb, adapter->netdev);
+		if (!rx_buffer_info->mapped_as_page)
+			skb->protocol = eth_type_trans(skb, adapter->netdev);
 #ifdef IXGBE_FCOE
 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
 		if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
@@ -984,7 +1063,14 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 				goto next_desc;
 		}
 #endif /* IXGBE_FCOE */
-		ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
+
+		if (!rx_buffer_info->mapped_as_page)
+			ixgbe_receive_skb(q_vector, skb, staterr,
+						rx_ring, rx_desc);
+		else {
+			skb_record_rx_queue(skb, rx_ring->queue_index);
+			napi_gro_frags(napi);
+		}
 
 next_desc:
 		rx_desc->wb.upper.status_error = 0;
@@ -3131,9 +3217,16 @@ static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
 
 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
 		if (rx_buffer_info->dma) {
-			pci_unmap_single(pdev, rx_buffer_info->dma,
-			                 rx_ring->rx_buf_len,
-			                 PCI_DMA_FROMDEVICE);
+			if (!rx_buffer_info->mapped_as_page) {
+				pci_unmap_single(pdev, rx_buffer_info->dma,
+						rx_ring->rx_buf_len,
+						PCI_DMA_FROMDEVICE);
+			} else {
+				pci_unmap_page(pdev, rx_buffer_info->dma,
+						PAGE_SIZE / 2,
+						PCI_DMA_FROMDEVICE);
+				rx_buffer_info->page_skb = NULL;
+			}
 			rx_buffer_info->dma = 0;
 		}
 		if (rx_buffer_info->skb) {
@@ -3158,7 +3251,7 @@ static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
 			               PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
 			rx_buffer_info->page_dma = 0;
 		}
-		put_page(rx_buffer_info->page);
+		netdev_free_page(adapter->netdev, rx_buffer_info->page);
 		rx_buffer_info->page = NULL;
 		rx_buffer_info->page_offset = 0;
 	}
-- 
1.5.4.4

[RFC PATCH v10 15/16] An example how to modifiy NIC driver to use napi_gro_frags() interface

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

This example is made on ixgbe driver.
It provides API is_rx_buffer_mapped_as_page() to indicate
if the driver use napi_gro_frags() interface or not.
The example allocates 2 pages for DMA for one ring descriptor
using netdev_alloc_page(). When packets is coming, using
napi_gro_frags() to allocate skb and to receive the packets.

---
 drivers/net/ixgbe/ixgbe.h      |    3 +
 drivers/net/ixgbe/ixgbe_main.c |  151 ++++++++++++++++++++++++++++++++--------
 2 files changed, 125 insertions(+), 29 deletions(-)

diff --git a/drivers/net/ixgbe/ixgbe.h b/drivers/net/ixgbe/ixgbe.h
index 79c35ae..fceffc5 100644
--- a/drivers/net/ixgbe/ixgbe.h
+++ b/drivers/net/ixgbe/ixgbe.h
@@ -131,6 +131,9 @@ struct ixgbe_rx_buffer {
 	struct page *page;
 	dma_addr_t page_dma;
 	unsigned int page_offset;
+	u16 mapped_as_page;
+	struct page *page_skb;
+	unsigned int page_skb_offset;
 };
 
 struct ixgbe_queue_stats {
diff --git a/drivers/net/ixgbe/ixgbe_main.c b/drivers/net/ixgbe/ixgbe_main.c
index 6c00ee4..905d6d2 100644
--- a/drivers/net/ixgbe/ixgbe_main.c
+++ b/drivers/net/ixgbe/ixgbe_main.c
@@ -688,6 +688,12 @@ static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
 	IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
 }
 
+static bool is_rx_buffer_mapped_as_page(struct ixgbe_rx_buffer *bi,
+					struct net_device *dev)
+{
+	return true;
+}
+
 /**
  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
  * @adapter: address of board private structure
@@ -704,13 +710,17 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 	i = rx_ring->next_to_use;
 	bi = &rx_ring->rx_buffer_info[i];
 
+
 	while (cleaned_count--) {
 		rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
 
+		bi->mapped_as_page =
+			is_rx_buffer_mapped_as_page(bi, adapter->netdev);
+
 		if (!bi->page_dma &&
 		    (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
 			if (!bi->page) {
-				bi->page = alloc_page(GFP_ATOMIC);
+				bi->page = netdev_alloc_page(adapter->netdev);
 				if (!bi->page) {
 					adapter->alloc_rx_page_failed++;
 					goto no_buffers;
@@ -727,7 +737,7 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 			                            PCI_DMA_FROMDEVICE);
 		}
 
-		if (!bi->skb) {
+		if (!bi->mapped_as_page && !bi->skb) {
 			struct sk_buff *skb;
 			/* netdev_alloc_skb reserves 32 bytes up front!! */
 			uint bufsz = rx_ring->rx_buf_len + SMP_CACHE_BYTES;
@@ -747,6 +757,19 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 			                         rx_ring->rx_buf_len,
 			                         PCI_DMA_FROMDEVICE);
 		}
+
+		if (bi->mapped_as_page && !bi->page_skb) {
+			bi->page_skb = netdev_alloc_page(adapter->netdev);
+			if (!bi->page_skb) {
+				adapter->alloc_rx_page_failed++;
+				goto no_buffers;
+			}
+			bi->page_skb_offset = 0;
+			bi->dma = pci_map_page(pdev, bi->page_skb,
+					bi->page_skb_offset,
+					(PAGE_SIZE / 2),
+					PCI_DMA_FROMDEVICE);
+		}
 		/* Refresh the desc even if buffer_addrs didn't change because
 		 * each write-back erases this info. */
 		if (rx_ring->flags & IXGBE_RING_RX_PS_ENABLED) {
@@ -823,6 +846,13 @@ struct ixgbe_rsc_cb {
 	dma_addr_t dma;
 };
 
+static bool is_no_buffer(struct ixgbe_rx_buffer *rx_buffer_info)
+{
+	return (!rx_buffer_info->skb ||
+		!rx_buffer_info->page_skb) &&
+		!rx_buffer_info->page;
+}
+
 #define IXGBE_RSC_CB(skb) ((struct ixgbe_rsc_cb *)(skb)->cb)
 
 static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
@@ -832,6 +862,7 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 	struct ixgbe_adapter *adapter = q_vector->adapter;
 	struct net_device *netdev = adapter->netdev;
 	struct pci_dev *pdev = adapter->pdev;
+	struct napi_struct *napi = &q_vector->napi;
 	union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
 	struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
 	struct sk_buff *skb;
@@ -868,29 +899,71 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 			len = le16_to_cpu(rx_desc->wb.upper.length);
 		}
 
+		if (is_no_buffer(rx_buffer_info))
+			break;
+
 		cleaned = true;
-		skb = rx_buffer_info->skb;
-		prefetch(skb->data);
-		rx_buffer_info->skb = NULL;
 
-		if (rx_buffer_info->dma) {
-			if ((adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) &&
-			    (!(staterr & IXGBE_RXD_STAT_EOP)) &&
-				 (!(skb->prev)))
-				/*
-				 * When HWRSC is enabled, delay unmapping
-				 * of the first packet. It carries the
-				 * header information, HW may still
-				 * access the header after the writeback.
-				 * Only unmap it when EOP is reached
-				 */
-				IXGBE_RSC_CB(skb)->dma = rx_buffer_info->dma;
-			else
-				pci_unmap_single(pdev, rx_buffer_info->dma,
-				                 rx_ring->rx_buf_len,
-				                 PCI_DMA_FROMDEVICE);
-			rx_buffer_info->dma = 0;
-			skb_put(skb, len);
+		if (!rx_buffer_info->mapped_as_page) {
+			skb = rx_buffer_info->skb;
+			prefetch(skb->data);
+			rx_buffer_info->skb = NULL;
+
+			if (rx_buffer_info->dma) {
+				if ((adapter->flags2 &
+					IXGBE_FLAG2_RSC_ENABLED) &&
+					(!(staterr & IXGBE_RXD_STAT_EOP)) &&
+					(!(skb->prev)))
+					/*
+					 * When HWRSC is enabled, delay unmapping
+					 * of the first packet. It carries the
+					 * header information, HW may still
+					 * access the header after the writeback.
+					 * Only unmap it when EOP is reached
+					 */
+					IXGBE_RSC_CB(skb)->dma =
+							rx_buffer_info->dma;
+				else
+					pci_unmap_single(pdev,
+							rx_buffer_info->dma,
+							rx_ring->rx_buf_len,
+							PCI_DMA_FROMDEVICE);
+				rx_buffer_info->dma = 0;
+				skb_put(skb, len);
+			}
+		} else {
+			skb = napi_get_frags(napi);
+			prefetch(rx_buffer_info->page_skb_offset);
+			rx_buffer_info->skb = NULL;
+			if (rx_buffer_info->dma) {
+				if ((adapter->flags2 &
+					IXGBE_FLAG2_RSC_ENABLED) &&
+					(!(staterr & IXGBE_RXD_STAT_EOP)) &&
+					(!(skb->prev)))
+					/*
+					 * When HWRSC is enabled, delay unmapping
+					 * of the first packet. It carries the
+					 * header information, HW may still
+					 * access the header after the writeback.
+					 * Only unmap it when EOP is reached
+					 */
+					IXGBE_RSC_CB(skb)->dma =
+						rx_buffer_info->dma;
+				else
+					pci_unmap_page(pdev, rx_buffer_info->dma,
+							PAGE_SIZE / 2,
+							PCI_DMA_FROMDEVICE);
+				rx_buffer_info->dma = 0;
+				skb_fill_page_desc(skb,
+						skb_shinfo(skb)->nr_frags,
+						rx_buffer_info->page_skb,
+						rx_buffer_info->page_skb_offset,
+						len);
+				rx_buffer_info->page_skb = NULL;
+				skb->len += len;
+				skb->data_len += len;
+				skb->truesize += len;
+			}
 		}
 
 		if (upper_len) {
@@ -956,6 +1029,11 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 				rx_buffer_info->dma = next_buffer->dma;
 				next_buffer->skb = skb;
 				next_buffer->dma = 0;
+				if (rx_buffer_info->mapped_as_page) {
+					rx_buffer_info->page_skb =
+							next_buffer->page_skb;
+					next_buffer->page_skb = NULL;
+				}
 			} else {
 				skb->next = next_buffer->skb;
 				skb->next->prev = skb;
@@ -975,7 +1053,8 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 		total_rx_bytes += skb->len;
 		total_rx_packets++;
 
-		skb->protocol = eth_type_trans(skb, adapter->netdev);
+		if (!rx_buffer_info->mapped_as_page)
+			skb->protocol = eth_type_trans(skb, adapter->netdev);
 #ifdef IXGBE_FCOE
 		/* if ddp, not passing to ULD unless for FCP_RSP or error */
 		if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED) {
@@ -984,7 +1063,14 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 				goto next_desc;
 		}
 #endif /* IXGBE_FCOE */
-		ixgbe_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
+
+		if (!rx_buffer_info->mapped_as_page)
+			ixgbe_receive_skb(q_vector, skb, staterr,
+						rx_ring, rx_desc);
+		else {
+			skb_record_rx_queue(skb, rx_ring->queue_index);
+			napi_gro_frags(napi);
+		}
 
 next_desc:
 		rx_desc->wb.upper.status_error = 0;
@@ -3131,9 +3217,16 @@ static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
 
 		rx_buffer_info = &rx_ring->rx_buffer_info[i];
 		if (rx_buffer_info->dma) {
-			pci_unmap_single(pdev, rx_buffer_info->dma,
-			                 rx_ring->rx_buf_len,
-			                 PCI_DMA_FROMDEVICE);
+			if (!rx_buffer_info->mapped_as_page) {
+				pci_unmap_single(pdev, rx_buffer_info->dma,
+						rx_ring->rx_buf_len,
+						PCI_DMA_FROMDEVICE);
+			} else {
+				pci_unmap_page(pdev, rx_buffer_info->dma,
+						PAGE_SIZE / 2,
+						PCI_DMA_FROMDEVICE);
+				rx_buffer_info->page_skb = NULL;
+			}
 			rx_buffer_info->dma = 0;
 		}
 		if (rx_buffer_info->skb) {
@@ -3158,7 +3251,7 @@ static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
 			               PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
 			rx_buffer_info->page_dma = 0;
 		}
-		put_page(rx_buffer_info->page);
+		netdev_free_page(adapter->netdev, rx_buffer_info->page);
 		rx_buffer_info->page = NULL;
 		rx_buffer_info->page_offset = 0;
 	}
-- 
1.5.4.4

[RFC PATCH v10 16/16]An example how to alloc user buffer based on napi_gro_frags() interface.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

This example is made on ixgbe driver which using napi_gro_frags().
It can get buffers from guest side directly using netdev_alloc_page()
and release guest buffers using netdev_free_page().

---
 drivers/net/ixgbe/ixgbe_main.c |   25 +++++++++++++++++++++----
 1 files changed, 21 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ixgbe/ixgbe_main.c b/drivers/net/ixgbe/ixgbe_main.c
index 905d6d2..0977f2f 100644
--- a/drivers/net/ixgbe/ixgbe_main.c
+++ b/drivers/net/ixgbe/ixgbe_main.c
@@ -691,7 +691,14 @@ static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
 static bool is_rx_buffer_mapped_as_page(struct ixgbe_rx_buffer *bi,
 					struct net_device *dev)
 {
-	return true;
+	return dev_is_mpassthru(dev);
+}
+
+static u32 get_page_skb_offset(struct net_device *dev)
+{
+	if (!dev_is_mpassthru(dev))
+		return 0;
+	return dev->mp_port->vnet_hlen;
 }
 
 /**
@@ -764,7 +771,8 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 				adapter->alloc_rx_page_failed++;
 				goto no_buffers;
 			}
-			bi->page_skb_offset = 0;
+			bi->page_skb_offset =
+				get_page_skb_offset(adapter->netdev);
 			bi->dma = pci_map_page(pdev, bi->page_skb,
 					bi->page_skb_offset,
 					(PAGE_SIZE / 2),
@@ -899,8 +907,10 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 			len = le16_to_cpu(rx_desc->wb.upper.length);
 		}
 
-		if (is_no_buffer(rx_buffer_info))
+		if (is_no_buffer(rx_buffer_info)) {
+			printk("no buffers\n");
 			break;
+		}
 
 		cleaned = true;
 
@@ -959,6 +969,12 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 						rx_buffer_info->page_skb,
 						rx_buffer_info->page_skb_offset,
 						len);
+				if (dev_is_mpassthru(netdev) &&
+						netdev->mp_port->hash)
+					skb_shinfo(skb)->destructor_arg =
+					netdev->mp_port->hash(netdev,
+					rx_buffer_info->page_skb);
+
 				rx_buffer_info->page_skb = NULL;
 				skb->len += len;
 				skb->data_len += len;
@@ -976,7 +992,8 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 			                   upper_len);
 
 			if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
-			    (page_count(rx_buffer_info->page) != 1))
+			    (page_count(rx_buffer_info->page) != 1) ||
+				dev_is_mpassthru(netdev))
 				rx_buffer_info->page = NULL;
 			else
 				get_page(rx_buffer_info->page);
-- 
1.5.4.4

[RFC PATCH v10 16/16]An example how to alloc user buffer based on napi_gro_frags() interface.

[xiaohui.xin]

From: Xin Xiaohui <xiaohui.xin@intel.com>

This example is made on ixgbe driver which using napi_gro_frags().
It can get buffers from guest side directly using netdev_alloc_page()
and release guest buffers using netdev_free_page().

---
 drivers/net/ixgbe/ixgbe_main.c |   25 +++++++++++++++++++++----
 1 files changed, 21 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ixgbe/ixgbe_main.c b/drivers/net/ixgbe/ixgbe_main.c
index 905d6d2..0977f2f 100644
--- a/drivers/net/ixgbe/ixgbe_main.c
+++ b/drivers/net/ixgbe/ixgbe_main.c
@@ -691,7 +691,14 @@ static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
 static bool is_rx_buffer_mapped_as_page(struct ixgbe_rx_buffer *bi,
 					struct net_device *dev)
 {
-	return true;
+	return dev_is_mpassthru(dev);
+}
+
+static u32 get_page_skb_offset(struct net_device *dev)
+{
+	if (!dev_is_mpassthru(dev))
+		return 0;
+	return dev->mp_port->vnet_hlen;
 }
 
 /**
@@ -764,7 +771,8 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 				adapter->alloc_rx_page_failed++;
 				goto no_buffers;
 			}
-			bi->page_skb_offset = 0;
+			bi->page_skb_offset =
+				get_page_skb_offset(adapter->netdev);
 			bi->dma = pci_map_page(pdev, bi->page_skb,
 					bi->page_skb_offset,
 					(PAGE_SIZE / 2),
@@ -899,8 +907,10 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 			len = le16_to_cpu(rx_desc->wb.upper.length);
 		}
 
-		if (is_no_buffer(rx_buffer_info))
+		if (is_no_buffer(rx_buffer_info)) {
+			printk("no buffers\n");
 			break;
+		}
 
 		cleaned = true;
 
@@ -959,6 +969,12 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 						rx_buffer_info->page_skb,
 						rx_buffer_info->page_skb_offset,
 						len);
+				if (dev_is_mpassthru(netdev) &&
+						netdev->mp_port->hash)
+					skb_shinfo(skb)->destructor_arg =
+					netdev->mp_port->hash(netdev,
+					rx_buffer_info->page_skb);
+
 				rx_buffer_info->page_skb = NULL;
 				skb->len += len;
 				skb->data_len += len;
@@ -976,7 +992,8 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
 			                   upper_len);
 
 			if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
-			    (page_count(rx_buffer_info->page) != 1))
+			    (page_count(rx_buffer_info->page) != 1) ||
+				dev_is_mpassthru(netdev))
 				rx_buffer_info->page = NULL;
 			else
 				get_page(rx_buffer_info->page);
-- 
1.5.4.4

RE: [RFC PATCH v10 00/16] Provide a zero-copy method on KVM virtio-net.

[Xin, Xiaohui]

Herbert,
Any comments on the modifications of the net core and driver side of this patch?

Thanks
Xiaohui

>-----Original Message-----
>From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-owner@vger.kernel.org] On
>Behalf Of xiaohui.xin@intel.com
>Sent: Saturday, September 11, 2010 5:53 PM
>To: netdev@vger.kernel.org; kvm@vger.kernel.org; linux-kernel@vger.kernel.org;
>mst@redhat.com; mingo@elte.hu; davem@davemloft.net; herbert@gondor.apana.org.au;
>jdike@linux.intel.com
>Subject: [RFC PATCH v10 00/16] Provide a zero-copy method on KVM virtio-net.
>
>We provide an zero-copy method which driver side may get external
>buffers to DMA. Here external means driver don't use kernel space
>to allocate skb buffers. Currently the external buffer can be from
>guest virtio-net driver.
>
>The idea is simple, just to pin the guest VM user space and then
>let host NIC driver has the chance to directly DMA to it.
>The patches are based on vhost-net backend driver. We add a device
>which provides proto_ops as sendmsg/recvmsg to vhost-net to
>send/recv directly to/from the NIC driver. KVM guest who use the
>vhost-net backend may bind any ethX interface in the host side to
>get copyless data transfer thru guest virtio-net frontend.
>
>patch 01-10:  	net core and kernel changes.
>patch 11-13:  	new device as interface to mantpulate external buffers.
>patch 14: 	for vhost-net.
>patch 15:	An example on modifying NIC driver to using napi_gro_frags().
>patch 16:	An example how to get guest buffers based on driver
>		who using napi_gro_frags().
>
>The guest virtio-net driver submits multiple requests thru vhost-net
>backend driver to the kernel. And the requests are queued and then
>completed after corresponding actions in h/w are done.
>
>For read, user space buffers are dispensed to NIC driver for rx when
>a page constructor API is invoked. Means NICs can allocate user buffers
>from a page constructor. We add a hook in netif_receive_skb() function
>to intercept the incoming packets, and notify the zero-copy device.
>
>For write, the zero-copy deivce may allocates a new host skb and puts
>payload on the skb_shinfo(skb)->frags, and copied the header to skb->data.
>The request remains pending until the skb is transmitted by h/w.
>
>We provide multiple submits and asynchronous notifiicaton to
>vhost-net too.
>
>Our goal is to improve the bandwidth and reduce the CPU usage.
>Exact performance data will be provided later.
>
>What we have not done yet:
>	Performance tuning
>
>what we have done in v1:
>	polish the RCU usage
>	deal with write logging in asynchroush mode in vhost
>	add notifier block for mp device
>	rename page_ctor to mp_port in netdevice.h to make it looks generic
>	add mp_dev_change_flags() for mp device to change NIC state
>	add CONIFG_VHOST_MPASSTHRU to limit the usage when module is not load
>	a small fix for missing dev_put when fail
>	using dynamic minor instead of static minor number
>	a __KERNEL__ protect to mp_get_sock()
>
>what we have done in v2:
>
>	remove most of the RCU usage, since the ctor pointer is only
>	changed by BIND/UNBIND ioctl, and during that time, NIC will be
>	stopped to get good cleanup(all outstanding requests are finished),
>	so the ctor pointer cannot be raced into wrong situation.
>
>	Remove the struct vhost_notifier with struct kiocb.
>	Let vhost-net backend to alloc/free the kiocb and transfer them
>	via sendmsg/recvmsg.
>
>	use get_user_pages_fast() and set_page_dirty_lock() when read.
>
>	Add some comments for netdev_mp_port_prep() and handle_mpassthru().
>
>what we have done in v3:
>	the async write logging is rewritten
>	a drafted synchronous write function for qemu live migration
>	a limit for locked pages from get_user_pages_fast() to prevent Dos
>	by using RLIMIT_MEMLOCK
>
>
>what we have done in v4:
>	add iocb completion callback from vhost-net to queue iocb in mp device
>	replace vq->receiver by mp_sock_data_ready()
>	remove stuff in mp device which access structures from vhost-net
>	modify skb_reserve() to ignore host NIC driver reserved space
>	rebase to the latest vhost tree
>	split large patches into small pieces, especially for net core part.
>
>
>what we have done in v5:
>	address Arnd Bergmann's comments
>		-remove IFF_MPASSTHRU_EXCL flag in mp device
>		-Add CONFIG_COMPAT macro
>		-remove mp_release ops
>	move dev_is_mpassthru() as inline func
>	fix a bug in memory relinquish
>	Apply to current git (2.6.34-rc6) tree.
>
>what we have done in v6:
>	move create_iocb() out of page_dtor which may happen in interrupt context
>	-This remove the potential issues which lock called in interrupt context
>	make the cache used by mp, vhost as static, and created/destoryed during
>	modules init/exit functions.
>	-This makes multiple mp guest created at the same time.
>
>what we have done in v7:
>	some cleanup prepared to suppprt PS mode
>
>what we have done in v8:
>	discarding the modifications to point skb->data to guest buffer directly.
>	Add code to modify driver to support napi_gro_frags() with Herbert's comments.
>	To support PS mode.
>	Add mergeable buffer support in mp device.
>	Add GSO/GRO support in mp deice.
>	Address comments from Eric Dumazet about cache line and rcu usage.
>
>what we have done in v9:
>	v8 patch is based on a fix in dev_gro_receive().
>	But Herbert did not agree with the fix we have sent out.
>	And he suggest another fix. v9 is modified to base on that fix.
>
>
>what we have done in v10:
>	Fix a partial csum error.
>	Cleanup some unused fields with struct page_info{} in mp device.
>	Modify kmem_cache_zalloc() to kmem_cache_alloc() based on Michael S. Thirkin.
>
>Performance:
>	We have seen the performance data request from mailling-list.
>	And we are now looking into this.
>--
>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>Please read the FAQ at  http://www.tux.org/lkml/

RE: [RFC PATCH v10 00/16] Provide a zero-copy method on KVM virtio-net.

[Xin, Xiaohui]

Herbert,
Any comments on the modifications of the net core and driver side of this patch?

Thanks
Xiaohui

>-----Original Message-----
>From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-owner@vger.kernel.org] On
>Behalf Of xiaohui.xin@intel.com
>Sent: Saturday, September 11, 2010 5:53 PM
>To: netdev@vger.kernel.org; kvm@vger.kernel.org; linux-kernel@vger.kernel.org;
>mst@redhat.com; mingo@elte.hu; davem@davemloft.net; herbert@gondor.apana.org.au;
>jdike@linux.intel.com
>Subject: [RFC PATCH v10 00/16] Provide a zero-copy method on KVM virtio-net.
>
>We provide an zero-copy method which driver side may get external
>buffers to DMA. Here external means driver don't use kernel space
>to allocate skb buffers. Currently the external buffer can be from
>guest virtio-net driver.
>
>The idea is simple, just to pin the guest VM user space and then
>let host NIC driver has the chance to directly DMA to it.
>The patches are based on vhost-net backend driver. We add a device
>which provides proto_ops as sendmsg/recvmsg to vhost-net to
>send/recv directly to/from the NIC driver. KVM guest who use the
>vhost-net backend may bind any ethX interface in the host side to
>get copyless data transfer thru guest virtio-net frontend.
>
>patch 01-10:  	net core and kernel changes.
>patch 11-13:  	new device as interface to mantpulate external buffers.
>patch 14: 	for vhost-net.
>patch 15:	An example on modifying NIC driver to using napi_gro_frags().
>patch 16:	An example how to get guest buffers based on driver
>		who using napi_gro_frags().
>
>The guest virtio-net driver submits multiple requests thru vhost-net
>backend driver to the kernel. And the requests are queued and then
>completed after corresponding actions in h/w are done.
>
>For read, user space buffers are dispensed to NIC driver for rx when
>a page constructor API is invoked. Means NICs can allocate user buffers
>from a page constructor. We add a hook in netif_receive_skb() function
>to intercept the incoming packets, and notify the zero-copy device.
>
>For write, the zero-copy deivce may allocates a new host skb and puts
>payload on the skb_shinfo(skb)->frags, and copied the header to skb->data.
>The request remains pending until the skb is transmitted by h/w.
>
>We provide multiple submits and asynchronous notifiicaton to
>vhost-net too.
>
>Our goal is to improve the bandwidth and reduce the CPU usage.
>Exact performance data will be provided later.
>
>What we have not done yet:
>	Performance tuning
>
>what we have done in v1:
>	polish the RCU usage
>	deal with write logging in asynchroush mode in vhost
>	add notifier block for mp device
>	rename page_ctor to mp_port in netdevice.h to make it looks generic
>	add mp_dev_change_flags() for mp device to change NIC state
>	add CONIFG_VHOST_MPASSTHRU to limit the usage when module is not load
>	a small fix for missing dev_put when fail
>	using dynamic minor instead of static minor number
>	a __KERNEL__ protect to mp_get_sock()
>
>what we have done in v2:
>
>	remove most of the RCU usage, since the ctor pointer is only
>	changed by BIND/UNBIND ioctl, and during that time, NIC will be
>	stopped to get good cleanup(all outstanding requests are finished),
>	so the ctor pointer cannot be raced into wrong situation.
>
>	Remove the struct vhost_notifier with struct kiocb.
>	Let vhost-net backend to alloc/free the kiocb and transfer them
>	via sendmsg/recvmsg.
>
>	use get_user_pages_fast() and set_page_dirty_lock() when read.
>
>	Add some comments for netdev_mp_port_prep() and handle_mpassthru().
>
>what we have done in v3:
>	the async write logging is rewritten
>	a drafted synchronous write function for qemu live migration
>	a limit for locked pages from get_user_pages_fast() to prevent Dos
>	by using RLIMIT_MEMLOCK
>
>
>what we have done in v4:
>	add iocb completion callback from vhost-net to queue iocb in mp device
>	replace vq->receiver by mp_sock_data_ready()
>	remove stuff in mp device which access structures from vhost-net
>	modify skb_reserve() to ignore host NIC driver reserved space
>	rebase to the latest vhost tree
>	split large patches into small pieces, especially for net core part.
>
>
>what we have done in v5:
>	address Arnd Bergmann's comments
>		-remove IFF_MPASSTHRU_EXCL flag in mp device
>		-Add CONFIG_COMPAT macro
>		-remove mp_release ops
>	move dev_is_mpassthru() as inline func
>	fix a bug in memory relinquish
>	Apply to current git (2.6.34-rc6) tree.
>
>what we have done in v6:
>	move create_iocb() out of page_dtor which may happen in interrupt context
>	-This remove the potential issues which lock called in interrupt context
>	make the cache used by mp, vhost as static, and created/destoryed during
>	modules init/exit functions.
>	-This makes multiple mp guest created at the same time.
>
>what we have done in v7:
>	some cleanup prepared to suppprt PS mode
>
>what we have done in v8:
>	discarding the modifications to point skb->data to guest buffer directly.
>	Add code to modify driver to support napi_gro_frags() with Herbert's comments.
>	To support PS mode.
>	Add mergeable buffer support in mp device.
>	Add GSO/GRO support in mp deice.
>	Address comments from Eric Dumazet about cache line and rcu usage.
>
>what we have done in v9:
>	v8 patch is based on a fix in dev_gro_receive().
>	But Herbert did not agree with the fix we have sent out.
>	And he suggest another fix. v9 is modified to base on that fix.
>
>
>what we have done in v10:
>	Fix a partial csum error.
>	Cleanup some unused fields with struct page_info{} in mp device.
>	Modify kmem_cache_zalloc() to kmem_cache_alloc() based on Michael S. Thirkin.
>
>Performance:
>	We have seen the performance data request from mailling-list.
>	And we are now looking into this.
>--
>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>Please read the FAQ at  http://www.tux.org/lkml/