From: Xin Xiaohui <[email protected]>
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
include/linux/netdevice.h | 19 ++++++++++++++++++-
1 files changed, 18 insertions(+), 1 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index c79a88b..bf79756 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -530,6 +530,22 @@ 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;
+ struct skb_external_page *(*ctor)(struct mpassthru_port *,
+ struct sk_buff *, int);
+};
/*
* This structure defines the management hooks for network devices.
@@ -952,7 +968,8 @@ 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
From: Xin Xiaohui <[email protected]>
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 <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
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 5c473fb..3a1583b 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -707,6 +707,10 @@ 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
From: Xin Xiaohui <[email protected]>
Make the skb->data and skb->head from external buffer
to be consistent, we ignore the room reserved by driver
for kernel skb.
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
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 8949b15..4458dc8 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -1206,6 +1206,15 @@ static inline int skb_tailroom(const struct sk_buff *skb)
*/
static inline void skb_reserve(struct sk_buff *skb, int len)
{
+ /* Since skb_reserve() is only for an empty buffer,
+ * and when the skb is getting external buffer, we cannot
+ * retain the external buffer has the same reserved space
+ * in the header which kernel allocatd skb has, so have to
+ * ignore this. And we have recorded the external buffer
+ * info in the destructor_arg field, so use it as indicator.
+ */
+ if (skb_shinfo(skb)->destructor_arg)
+ return;
skb->data += len;
skb->tail += len;
}
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
include/linux/mpassthru.h | 29 +++++++++++++++++++++++++++++
1 files changed, 29 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..e3983d3
--- /dev/null
+++ b/include/linux/mpassthru.h
@@ -0,0 +1,29 @@
+#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 _IOW('M', 214, int)
+
+/* MPASSTHRU ifc flags */
+#define IFF_MPASSTHRU 0x0001
+#define IFF_MPASSTHRU_EXCL 0x0002
+
+#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
From: Xin Xiaohui <[email protected]>
How external buffer comes from, how to destroy.
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
drivers/vhost/mpassthru.c | 237 ++++++++++++++++++++++++++++++++++++++++++++-
1 files changed, 236 insertions(+), 1 deletions(-)
diff --git a/drivers/vhost/mpassthru.c b/drivers/vhost/mpassthru.c
index c5ede17..b171f21 100644
--- a/drivers/vhost/mpassthru.c
+++ b/drivers/vhost/mpassthru.c
@@ -159,6 +159,39 @@ static int mp_dev_change_flags(struct net_device *dev, unsigned flags)
return ret;
}
+/* The main function to allocate external buffers */
+static struct skb_external_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);
+ }
+ spin_unlock_irqrestore(&ctor->read_lock, flags);
+ if (!info)
+ return NULL;
+
+ for (i = 0; i < info->pnum; i++) {
+ get_page(info->pages[i]);
+ info->frag[i].page = info->pages[i];
+ info->frag[i].page_offset = i ? 0 : info->offset;
+ info->frag[i].size = port->npages > 1 ? PAGE_SIZE :
+ port->data_len;
+ }
+ info->skb = skb;
+ info->ext_page.frags = info->frag;
+ info->ext_page.ushinfo = &info->ushinfo;
+ return &info->ext_page;
+}
+
static int page_ctor_attach(struct mp_struct *mp)
{
int rc;
@@ -191,7 +224,7 @@ static int page_ctor_attach(struct mp_struct *mp)
dev_hold(dev);
ctor->dev = dev;
- ctor->port.ctor = NULL;
+ ctor->port.ctor = page_ctor;
ctor->port.sock = &mp->socket;
ctor->lock_pages = 0;
rc = netdev_mp_port_attach(dev, &ctor->port);
@@ -258,6 +291,52 @@ static int set_memlock_rlimit(struct page_ctor *ctor, int resource,
task_unlock(current->group_leader);
return 0;
}
+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]);
+ }
+ }
+ skb_shinfo(info->skb)->destructor_arg = &info->ext_page;
+ info->skb->destructor = NULL;
+ kfree_skb(info->skb);
+ }
+ /* Decrement the number of locked pages */
+ info->ctor->lock_pages -= info->pnum;
+ kmem_cache_free(info->ctor->cache, info);
+
+ 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_iovec = 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)
{
@@ -275,6 +354,7 @@ static int page_ctor_detach(struct mp_struct *mp)
for (i = 0; i < info->pnum; i++)
if (info->pages[i])
put_page(info->pages[i]);
+ create_iocb(info, 0);
kmem_cache_free(ctor->cache, info);
}
set_memlock_rlimit(ctor, RLIMIT_MEMLOCK,
@@ -328,6 +408,161 @@ static void mp_put(struct mp_file *mfile)
mp_detach(mfile->mp);
}
+/* The callback to destruct the external buffers or skb */
+static void page_dtor(struct skb_external_page *ext_page)
+{
+ struct page_info *info;
+ struct page_ctor *ctor;
+ struct sock *sk;
+ struct sk_buff *skb;
+ struct kiocb *iocb = NULL;
+ unsigned long flags;
+
+ 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) && info->skb)
+ info->skb->head = NULL;
+
+ /* If the info->total is 0, make it to be reused */
+ if (!info->total) {
+ spin_lock_irqsave(&ctor->read_lock, flags);
+ list_add(&info->list, &ctor->readq);
+ spin_unlock_irqrestore(&ctor->read_lock, flags);
+ return;
+ }
+
+ if (info->flags == INFO_READ)
+ return;
+
+ /* For transmit, we should wait for the DMA finish by hardware.
+ * Queue the notifier to wake up the backend driver
+ */
+
+ iocb = create_iocb(info, info->total);
+
+ 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_zalloc(ctor->cache, GFP_KERNEL);
+
+ if (!info)
+ return NULL;
+ info->total = total;
+ info->ext_page.dtor = page_dtor;
+ info->ctor = ctor;
+ info->flags = INFO_WRITE;
+ info->iocb = iocb;
+ return info;
+}
+
+/* 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) {
+ printk(KERN_INFO "exceed the locked memory rlimit.");
+ return NULL;
+ }
+
+ info = kmem_cache_zalloc(ctor->cache, GFP_KERNEL);
+
+ if (!info)
+ return 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->total = total;
+ info->ext_page.dtor = page_dtor;
+ info->ctor = ctor;
+ info->pnum = j;
+ info->iocb = iocb;
+ if (!npages)
+ info->flags = INFO_WRITE;
+ if (info->flags == INFO_READ) {
+ info->ext_page.start = (u8 *)(((unsigned long)
+ (pfn_to_kaddr(page_to_pfn(info->pages[0]))) +
+ frags[0].offset));
+#ifdef NET_SKBUFF_DATA_USES_OFFSET
+ info->ext_page.size = SKB_DATA_ALIGN(
+ iov[0].iov_len + NET_IP_ALIGN + NET_SKB_PAD);
+#else
+ info->ext_page.size = SKB_DATA_ALIGN(
+ iov[0].iov_len + NET_IP_ALIGN + NET_SKB_PAD) -
+ NET_IP_ALIGN - NET_SKB_PAD;
+#endif
+ }
+ /* 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(ctor->cache, info);
+
+ return NULL;
+}
+
/* Ops structure to mimic raw sockets with mp device */
static const struct proto_ops mp_socket_ops = {
};
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
Provides multiple submits and asynchronous notifications.
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 <[email protected]>
---
drivers/vhost/net.c | 236 +++++++++++++++++++++++++++++++++++++++++++++++-
drivers/vhost/vhost.c | 120 ++++++++++++++-----------
drivers/vhost/vhost.h | 14 +++
3 files changed, 314 insertions(+), 56 deletions(-)
diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index 38989d1..18f6c41 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -23,6 +23,8 @@
#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>
@@ -48,6 +50,7 @@ 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. */
@@ -92,11 +95,138 @@ 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;
+
+ 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) {
+ 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 info is needed,
+ * since these buffers are in async queue, and may not get
+ * the log info before.
+ */
+ if (unlikely(vq_log)) {
+ if (!log)
+ __vhost_get_vq_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;
+ }
+ }
+}
+
+static void handle_async_tx_events_notify(struct vhost_net *net,
+ struct vhost_virtqueue *vq)
+{
+ struct kiocb *iocb = NULL;
+ int tx_total_len = 0;
+
+ if (!is_async_vq(vq))
+ return;
+
+ while ((iocb = notify_dequeue(vq)) != NULL) {
+ 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;
+ }
+ }
+}
+
+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,
@@ -129,6 +259,8 @@ static void handle_tx(struct vhost_net *net)
tx_poll_stop(net);
hdr_size = vq->hdr_size;
+ handle_async_tx_events_notify(net, vq);
+
for (;;) {
head = vhost_get_vq_desc(&net->dev, vq, vq->iov,
ARRAY_SIZE(vq->iov),
@@ -156,6 +288,13 @@ static void handle_tx(struct vhost_net *net)
/* Skip header. TODO: support TSO. */
s = move_iovec_hdr(vq->iov, vq->hdr, hdr_size, out);
msg.msg_iovlen = out;
+
+ if (is_async_vq(vq)) {
+ iocb = create_iocb(net, vq, head);
+ if (!iocb)
+ break;
+ }
+
len = iov_length(vq->iov, out);
/* Sanity check */
if (!len) {
@@ -165,12 +304,18 @@ 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_vq_desc(vq);
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);
@@ -182,6 +327,8 @@ static void handle_tx(struct vhost_net *net)
}
}
+ handle_async_tx_events_notify(net, vq);
+
mutex_unlock(&vq->mutex);
unuse_mm(net->dev.mm);
}
@@ -191,6 +338,7 @@ static void handle_tx(struct vhost_net *net)
static void handle_rx(struct vhost_net *net)
{
struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX];
+ struct kiocb *iocb = NULL;
unsigned head, out, in, log, s;
struct vhost_log *vq_log;
struct msghdr msg = {
@@ -211,7 +359,8 @@ static void handle_rx(struct vhost_net *net)
int err;
size_t hdr_size;
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) &&
+ vq->link_state == VHOST_VQ_LINK_SYNC))
return;
use_mm(net->dev.mm);
@@ -219,9 +368,17 @@ static void handle_rx(struct vhost_net *net)
vhost_disable_notify(vq);
hdr_size = vq->hdr_size;
+ /* 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;
+ handle_async_rx_events_notify(net, vq, sock);
+
for (;;) {
head = vhost_get_vq_desc(&net->dev, vq, vq->iov,
ARRAY_SIZE(vq->iov),
@@ -250,6 +407,13 @@ static void handle_rx(struct vhost_net *net)
s = move_iovec_hdr(vq->iov, vq->hdr, hdr_size, in);
msg.msg_iovlen = in;
len = iov_length(vq->iov, in);
+
+ 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 RX: "
@@ -257,13 +421,20 @@ static void handle_rx(struct vhost_net *net)
iov_length(vq->hdr, s), hdr_size);
break;
}
- err = sock->ops->recvmsg(NULL, sock, &msg,
+
+ 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_vq_desc(vq);
break;
}
+
+ if (is_async_vq(vq))
+ continue;
+
/* TODO: Should check and handle checksum. */
if (err > len) {
pr_err("Discarded truncated rx packet: "
@@ -289,6 +460,8 @@ 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);
}
@@ -342,6 +515,7 @@ 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;
@@ -405,6 +579,18 @@ 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 = &n->dev.vqs[VHOST_NET_VQ_RX];
+ struct kiocb *iocb = NULL;
+ if (n->cache) {
+ while ((iocb = notify_dequeue(vq)) != NULL)
+ kmem_cache_free(n->cache, iocb);
+ kmem_cache_destroy(n->cache);
+ }
+}
+
static int vhost_net_release(struct inode *inode, struct file *f)
{
struct vhost_net *n = f->private_data;
@@ -421,6 +607,7 @@ 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;
}
@@ -472,21 +659,58 @@ 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;
+ 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 = kmem_cache_create("vhost_kiocb",
+ sizeof(struct kiocb), 0,
+ SLAB_HWCACHE_ALIGN, NULL);
+ }
+ }
+}
+
static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
{
struct socket *sock, *oldsock;
@@ -510,12 +734,14 @@ 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)
diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 3f10194..b39e47c 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -860,61 +860,17 @@ static unsigned get_indirect(struct vhost_dev *dev, struct vhost_virtqueue *vq,
return 0;
}
-/* 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
- * iovecs, but we pack them into one and note how many of each there were.
- *
- * This function returns the descriptor number found, or vq->num (which
- * is never a valid descriptor number) if none was found. */
-unsigned vhost_get_vq_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)
+/* This computes the log info according to the index of buffer */
+unsigned __vhost_get_vq_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, head, found = 0;
- u16 last_avail_idx;
+ unsigned int i = head, found = 0;
int ret;
- /* Check it isn't doing very strange things with descriptor numbers. */
- last_avail_idx = vq->last_avail_idx;
- if (get_user(vq->avail_idx, &vq->avail->idx)) {
- vq_err(vq, "Failed to access avail idx at %p\n",
- &vq->avail->idx);
- return vq->num;
- }
-
- if ((u16)(vq->avail_idx - last_avail_idx) > vq->num) {
- vq_err(vq, "Guest moved used index from %u to %u",
- last_avail_idx, vq->avail_idx);
- return vq->num;
- }
-
- /* If there's nothing new since last we looked, return invalid. */
- if (vq->avail_idx == last_avail_idx)
- return vq->num;
-
- /* Only get avail ring entries after they have been exposed by guest. */
- smp_rmb();
-
- /* Grab the next descriptor number they're advertising, and increment
- * the index we've seen. */
- if (get_user(head, &vq->avail->ring[last_avail_idx % vq->num])) {
- vq_err(vq, "Failed to read head: idx %d address %p\n",
- last_avail_idx,
- &vq->avail->ring[last_avail_idx % vq->num]);
- return vq->num;
- }
-
- /* If their number is silly, that's an error. */
- if (head >= vq->num) {
- vq_err(vq, "Guest says index %u > %u is available",
- head, vq->num);
- return vq->num;
- }
-
- /* When we start there are none of either input nor output. */
*out_num = *in_num = 0;
if (unlikely(log))
*log_num = 0;
@@ -978,8 +934,70 @@ unsigned vhost_get_vq_desc(struct vhost_dev *dev, struct vhost_virtqueue *vq,
*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
+ * iovecs, but we pack them into one and note how many of each there were.
+ *
+ * This function returns the descriptor number found, or vq->num (which
+ * is never a valid descriptor number) if none was found. */
+unsigned vhost_get_vq_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)
+{
+ struct vring_desc desc;
+ unsigned int i, head, found = 0;
+ u16 last_avail_idx;
+ int ret;
+
+ /* Check it isn't doing very strange things with descriptor numbers. */
+ last_avail_idx = vq->last_avail_idx;
+ if (get_user(vq->avail_idx, &vq->avail->idx)) {
+ vq_err(vq, "Failed to access avail idx at %p\n",
+ &vq->avail->idx);
+ return vq->num;
+ }
+
+ if ((u16)(vq->avail_idx - last_avail_idx) > vq->num) {
+ vq_err(vq, "Guest moved used index from %u to %u",
+ last_avail_idx, vq->avail_idx);
+ return vq->num;
+ }
+
+ /* If there's nothing new since last we looked, return invalid. */
+ if (vq->avail_idx == last_avail_idx)
+ return vq->num;
+
+ /* Only get avail ring entries after they have been exposed by guest. */
+ smp_rmb();
+
+ /* Grab the next descriptor number they're advertising, and increment
+ * the index we've seen. */
+ if (get_user(head, &vq->avail->ring[last_avail_idx % vq->num])) {
+ vq_err(vq, "Failed to read head: idx %d address %p\n",
+ last_avail_idx,
+ &vq->avail->ring[last_avail_idx % vq->num]);
+ return vq->num;
+ }
+
+ /* If their number is silly, that's an error. */
+ if (head >= vq->num) {
+ vq_err(vq, "Guest says index %u > %u is available",
+ head, vq->num);
+ return vq->num;
+ }
+
+ ret = __vhost_get_vq_desc(dev, vq, iov, iov_size,
+ out_num, in_num,
+ log, log_num, head);
/* On success, increment avail index. */
+ if (ret == vq->num)
+ return ret;
vq->last_avail_idx++;
return head;
}
diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index 44591ba..3c9cbce 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;
@@ -96,6 +101,10 @@ 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 {
@@ -124,6 +133,11 @@ unsigned vhost_get_vq_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 __vhost_get_vq_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);
void vhost_discard_vq_desc(struct vhost_virtqueue *);
int vhost_add_used(struct vhost_virtqueue *, unsigned int head, int len);
--
1.5.4.4
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 01-12: net core changes.
patch 13-17: new device as interface to mantpulate external buffers.
patch 18: for vhost-net.
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.
Here, we have ever considered 2 ways to utilize the page constructor
API to dispense the user buffers.
One: Modify __alloc_skb() function a bit, it can only allocate a
structure of sk_buff, and the data pointer is pointing to a
user buffer which is coming from a page constructor API.
Then the shinfo of the skb is also from guest.
When packet is received from hardware, the skb->data is filled
directly by h/w. What we have done is in this way.
Pros: We can avoid any copy here.
Cons: Guest virtio-net driver needs to allocate skb as almost
the same method with the host NIC drivers, say the size
of netdev_alloc_skb() and the same reserved space in the
head of skb. Many NIC drivers are the same with guest and
ok for this. But some lastest NIC drivers reserves special
room in skb head. To deal with it, we suggest to provide
a method in guest virtio-net driver to ask for parameter
we interest from the NIC driver when we know which device
we have bind to do zero-copy. Then we ask guest to do so.
Is that reasonable?
Two: Modify driver to get user buffer allocated from a page constructor
API(to substitute alloc_page()), the user buffer are used as payload
buffers and filled by h/w directly when packet is received. Driver
should associate the pages with skb (skb_shinfo(skb)->frags). For
the head buffer side, let host allocates skb, and h/w fills it.
After that, the data filled in host skb header will be copied into
guest header buffer which is submitted together with the payload buffer.
Pros: We could less care the way how guest or host allocates their
buffers.
Cons: We still need a bit copy here for the skb header.
We are not sure which way is the better here. This is the first thing we want
to get comments from the community. We wish the modification to the network
part will be generic which not used by vhost-net backend only, but a user
application may use it as well when the zero-copy device may provides async
read/write operations later.
Please give comments especially for the network part modifications.
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. But for simple
test with netperf, we found bindwidth up and CPU % up too,
but the bindwidth up ratio is much more than CPU % up ratio.
What we have not done yet:
packet split support
To support GRO
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.
performance:
using netperf with GSO/TSO disabled, 10G NIC,
disabled packet split mode, with raw socket case compared to vhost.
bindwidth will be from 1.1Gbps to 1.7Gbps
CPU % from 120%-140% to 140%-160%
From: Xin Xiaohui <[email protected]>
Currently, it can get external buffers from mp device.
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
include/linux/skbuff.h | 12 ++++++++++++
net/core/skbuff.c | 16 ++++++++++++++++
2 files changed, 28 insertions(+), 0 deletions(-)
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 3104e7d..96799f5 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -1525,6 +1525,18 @@ static inline void netdev_free_page(struct net_device *dev, struct page *page)
__free_page(page);
}
+extern struct skb_external_page *netdev_alloc_external_pages(
+ struct net_device *dev,
+ struct sk_buff *skb, int npages);
+
+static inline struct skb_external_page *netdev_alloc_external_page(
+ struct net_device *dev,
+ struct sk_buff *skb, unsigned int size)
+{
+ return netdev_alloc_external_pages(dev, skb,
+ DIV_ROUND_UP(size, PAGE_SIZE));
+}
+
/**
* skb_clone_writable - is the header of a clone writable
* @skb: buffer to check
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 93c4e06..6345acc 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -278,6 +278,22 @@ struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
}
EXPORT_SYMBOL(__netdev_alloc_page);
+struct skb_external_page *netdev_alloc_external_pages(struct net_device *dev,
+ struct sk_buff *skb, int npages)
+{
+ struct mpassthru_port *port;
+ struct skb_external_page *ext_page = NULL;
+
+ port = rcu_dereference(dev->mp_port);
+ if (!port)
+ goto out;
+ BUG_ON(npages > port->npages);
+ ext_page = port->ctor(port, skb, npages);
+out:
+ return ext_page;
+}
+EXPORT_SYMBOL(netdev_alloc_external_pages);
+
void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
int size)
{
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
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 ae223d2..169f22c 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -553,6 +553,12 @@ 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
From: Xin Xiaohui <[email protected]>
The ioctl is used by mp device to bind an underlying
NIC, it will query hardware capability and declare the
NIC to use external buffers.
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
memory leak fixed,
kconfig made,
do_unbind() made,
mp_chr_ioctl() cleanup
by Jeff Dike <[email protected]>
drivers/vhost/mpassthru.c | 671 +++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 671 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..c5ede17
--- /dev/null
+++ b/drivers/vhost/mpassthru.c
@@ -0,0 +1,671 @@
+/*
+ * 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/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;
+};
+
+struct page_info {
+ struct list_head list;
+ int header;
+ /* indicate the actual length of bytes
+ * send/recv in the external buffers
+ */
+ int total;
+ int offset;
+ struct page *pages[MAX_SKB_FRAGS+1];
+ struct skb_frag_struct frag[MAX_SKB_FRAGS+1];
+ 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_external_page ext_page;
+ struct skb_shared_info ushinfo;
+
+#define INFO_READ 0
+#define INFO_WRITE 1
+ unsigned flags;
+ unsigned pnum;
+
+ /* It's meaningful for receive, means
+ * the max length allowed
+ */
+ size_t len;
+
+ /* The fields after that is for backend
+ * driver, now for vhost-net.
+ */
+
+ struct kiocb *iocb;
+ unsigned int desc_pos;
+ struct iovec hdr[MAX_SKB_FRAGS + 2];
+ struct iovec iov[MAX_SKB_FRAGS + 2];
+};
+
+struct page_ctor {
+ struct list_head readq;
+ int w_len;
+ int r_len;
+ spinlock_t read_lock;
+ struct kmem_cache *cache;
+ /* record the locked pages */
+ int lock_pages;
+ struct rlimit o_rlim;
+ struct net_device *dev;
+ struct mpassthru_port port;
+};
+
+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;
+}
+
+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 (rcu_dereference(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;
+
+ ctor->cache = kmem_cache_create("skb_page_info",
+ sizeof(struct page_info), 0,
+ SLAB_HWCACHE_ALIGN, NULL);
+
+ if (!ctor->cache)
+ goto cache_fail;
+
+ INIT_LIST_HEAD(&ctor->readq);
+ spin_lock_init(&ctor->read_lock);
+
+ ctor->w_len = 0;
+ ctor->r_len = 0;
+
+ dev_hold(dev);
+ ctor->dev = dev;
+ ctor->port.ctor = NULL;
+ ctor->port.sock = &mp->socket;
+ ctor->lock_pages = 0;
+ rc = netdev_mp_port_attach(dev, &ctor->port);
+ if (rc)
+ goto fail;
+
+ /* locked by mp_mutex */
+ rcu_assign_pointer(mp->ctor, ctor);
+
+ /* XXX:Need we do set_offload here ? */
+
+ return 0;
+
+fail:
+ kmem_cache_destroy(ctor->cache);
+cache_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);
+ }
+ 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 int page_ctor_detach(struct mp_struct *mp)
+{
+ struct page_ctor *ctor;
+ struct page_info *info;
+ struct kiocb *iocb = NULL;
+ int i;
+
+ /* locked by mp_mutex */
+ ctor = rcu_dereference(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]);
+ kmem_cache_free(ctor->cache, info);
+ }
+ set_memlock_rlimit(ctor, RLIMIT_MEMLOCK,
+ ctor->o_rlim.rlim_cur,
+ ctor->o_rlim.rlim_max);
+ kmem_cache_destroy(ctor->cache);
+ netdev_mp_port_detach(ctor->dev);
+ dev_put(ctor->dev);
+
+ /* locked by mp_mutex */
+ rcu_assign_pointer(mp->ctor, NULL);
+ synchronize_rcu();
+
+ 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);
+}
+
+/* Ops structure to mimic raw sockets with mp device */
+static const struct proto_ops mp_socket_ops = {
+};
+
+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 = -EBUSY;
+
+ if (ifr.ifr_flags & IFF_MPASSTHRU_EXCL)
+ break;
+
+ ret = -ENODEV;
+ dev = dev_get_by_name(mfile->net, ifr.ifr_name);
+ if (!dev)
+ break;
+
+ mutex_lock(&mp_mutex);
+
+ ret = -EBUSY;
+ 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 = NULL;
+ sk->sk_data_ready = NULL;
+ sk->sk_write_space = NULL;
+ sk->sk_state_change = NULL;
+ ret = mp_attach(mp, file);
+ if (ret < 0)
+ goto err_free_sk;
+
+ ret = page_ctor_attach(mp);
+ if (ret < 0)
+ goto err_free_sk;
+
+ ifr.ifr_flags |= IFF_MPASSTHRU_EXCL;
+ mp_dev_change_flags(mp->dev, mp->dev->flags | IFF_UP);
+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;
+}
+
+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,
+ .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;
+
+ 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;
+ }
+ return NOTIFY_DONE;
+}
+
+static struct notifier_block mp_notifier_block __read_mostly = {
+ .notifier_call = mp_device_event,
+};
+
+static int mp_init(void)
+{
+ int ret = 0;
+
+ ret = misc_register(&mp_miscdev);
+ if (ret)
+ printk(KERN_ERR "mp: Can't register misc device\n");
+ else {
+ printk(KERN_INFO "Registering mp misc device - minor = %d\n",
+ mp_miscdev.minor);
+ register_netdevice_notifier(&mp_notifier_block);
+ }
+ return ret;
+}
+
+void mp_cleanup(void)
+{
+ unregister_netdevice_notifier(&mp_notifier_block);
+ misc_deregister(&mp_miscdev);
+}
+
+/* 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_cleanup);
+MODULE_AUTHOR(DRV_COPYRIGHT);
+MODULE_DESCRIPTION(DRV_DESCRIPTION);
+MODULE_LICENSE("GPL v2");
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
net/core/dev.c | 4 ++++
1 files changed, 4 insertions(+), 0 deletions(-)
diff --git a/net/core/dev.c b/net/core/dev.c
index 3e3ad23..f275c44 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2780,6 +2780,10 @@ enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
if (skb_is_gso(skb) || skb_has_frags(skb))
goto normal;
+ /* currently GRO is not supported by mediate passthru */
+ if (dev_is_mpassthru(skb->dev))
+ goto normal;
+
rcu_read_lock();
list_for_each_entry_rcu(ptype, head, list) {
if (ptype->type != type || ptype->dev || !ptype->gro_receive)
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
Currently, vhost-net is only user to the mp device.
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
drivers/vhost/mpassthru.c | 321 ++++++++++++++++++++++++++++++++++++++++++++-
1 files changed, 317 insertions(+), 4 deletions(-)
diff --git a/drivers/vhost/mpassthru.c b/drivers/vhost/mpassthru.c
index b171f21..0ac1a71 100644
--- a/drivers/vhost/mpassthru.c
+++ b/drivers/vhost/mpassthru.c
@@ -563,8 +563,321 @@ failed:
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;
+ struct ethhdr *eth;
+ struct kiocb *iocb = NULL;
+ int len, i;
+
+ struct virtio_net_hdr hdr = {
+ .flags = 0,
+ .gso_type = VIRTIO_NET_HDR_GSO_NONE
+ };
+
+ ctor = rcu_dereference(mp->ctor);
+ if (!ctor)
+ return;
+
+ while ((skb = skb_dequeue(&sk->sk_receive_queue)) != NULL) {
+ if (skb_shinfo(skb)->destructor_arg) {
+ info = container_of(skb_shinfo(skb)->destructor_arg,
+ struct page_info, ext_page);
+ info->skb = skb;
+ if (skb->len > info->len) {
+ mp->dev->stats.rx_dropped++;
+ DBG(KERN_INFO "Discarded truncated rx packet: "
+ " len %d > %zd\n", skb->len, info->len);
+ info->total = skb->len;
+ goto clean;
+ } else {
+ int i;
+ struct skb_shared_info *gshinfo =
+ (struct skb_shared_info *)
+ (&info->ushinfo);
+ struct skb_shared_info *hshinfo =
+ skb_shinfo(skb);
+
+ if (gshinfo->nr_frags < hshinfo->nr_frags)
+ goto clean;
+ eth = eth_hdr(skb);
+ skb_push(skb, ETH_HLEN);
+
+ hdr.hdr_len = skb_headlen(skb);
+ info->total = skb->len;
+
+ for (i = 0; i < gshinfo->nr_frags; i++)
+ gshinfo->frags[i].size = 0;
+ for (i = 0; i < hshinfo->nr_frags; i++)
+ gshinfo->frags[i].size =
+ hshinfo->frags[i].size;
+ memcpy(skb_shinfo(skb), &info->ushinfo,
+ sizeof(struct skb_shared_info));
+ }
+ } else {
+ /* The skb composed with kernel buffers
+ * in case external buffers are not sufficent.
+ * The case should be rare.
+ */
+ unsigned long flags;
+ int i;
+ struct skb_shared_info *gshinfo = NULL;
+
+ 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);
+ }
+ spin_unlock_irqrestore(&ctor->read_lock, flags);
+ if (!info) {
+ DBG(KERN_INFO
+ "No external buffer avaliable %p\n",
+ skb);
+ skb_queue_head(&sk->sk_receive_queue,
+ skb);
+ break;
+ }
+ info->skb = skb;
+ /* compute the guest skb frags info */
+ gshinfo = (struct skb_shared_info *)
+ (info->ext_page.start +
+ SKB_DATA_ALIGN(info->ext_page.size));
+
+ if (gshinfo->nr_frags < skb_shinfo(skb)->nr_frags)
+ goto clean;
+
+ eth = eth_hdr(skb);
+ skb_push(skb, ETH_HLEN);
+ info->total = skb->len;
+
+ for (i = 0; i < gshinfo->nr_frags; i++)
+ gshinfo->frags[i].size = 0;
+ for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
+ gshinfo->frags[i].size =
+ skb_shinfo(skb)->frags[i].size;
+ hdr.hdr_len = min_t(int, skb->len,
+ info->iov[1].iov_len);
+ skb_copy_datagram_iovec(skb, 0, info->iov, skb->len);
+ }
+
+ len = memcpy_toiovec(info->hdr, (unsigned char *)&hdr,
+ sizeof hdr);
+ if (len) {
+ DBG(KERN_INFO
+ "Unable to write vnet_hdr at addr %p: %d\n",
+ info->hdr->iov_base, len);
+ goto clean;
+ }
+
+ iocb = create_iocb(info, skb->len + sizeof(hdr));
+ continue;
+
+clean:
+ kfree_skb(skb);
+ for (i = 0; info->pages[i]; i++)
+ put_page(info->pages[i]);
+ kmem_cache_free(ctor->cache, info);
+ }
+ return;
+}
+
+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 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 = rcu_dereference(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:
+ header = total > COPY_THRESHOLD ? COPY_HDR_LEN : total;
+
+ skb = alloc_skb(header + NET_IP_ALIGN, GFP_ATOMIC);
+ if (!skb)
+ goto drop;
+
+ skb_reserve(skb, NET_IP_ALIGN);
+
+ skb_set_network_header(skb, ETH_HLEN);
+
+ memcpy_fromiovec(skb->data, iov, header);
+ skb_put(skb, header);
+ skb->protocol = *((__be16 *)(skb->data) + ETH_ALEN);
+
+ 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; info->pages[i]; 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->total = total;
+ info->skb = skb;
+ skb_shinfo(skb)->destructor_arg = &info->ext_page;
+ skb->dev = mp->dev;
+ dev_queue_xmit(skb);
+ return 0;
+ }
+drop:
+ kfree_skb(skb);
+ if (info) {
+ for (i = 0; info->pages[i]; i++)
+ put_page(info->pages[i]);
+ kmem_cache_free(info->ctor->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 = rcu_dereference(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 */
+ iov++;
+ count--;
+
+ if (!ctor->lock_pages)
+ set_memlock_rlimit(ctor, RLIMIT_MEMLOCK,
+ iocb->ki_user_data * 4096,
+ iocb->ki_user_data * 4096);
+
+ /* Translate address to kernel */
+ info = alloc_page_info(ctor, iocb, iov, count, frags, npages, 0);
+ if (!info)
+ return -ENOMEM;
+ info->len = total_len;
+ info->hdr[0].iov_base = iocb->ki_iovec[0].iov_base;
+ info->hdr[0].iov_len = iocb->ki_iovec[0].iov_len;
+ info->offset = frags[0].offset;
+ info->desc_pos = iocb->ki_pos;
+
+ 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);
+
+ 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 = {
@@ -687,10 +1000,10 @@ static long mp_chr_ioctl(struct file *file, unsigned int cmd,
sk->sk_sndbuf = INT_MAX;
container_of(sk, struct mp_sock, sk)->mp = mp;
- sk->sk_destruct = NULL;
- sk->sk_data_ready = NULL;
- sk->sk_write_space = NULL;
- sk->sk_state_change = NULL;
+ 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;
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
Signed-off-by: Xin Xiaohui <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
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
From: Xin Xiaohui <[email protected]>
Add a dev parameter to __alloc_skb(), skb->data
points to external buffer, recompute skb->head,
maintain shinfo of the external buffer, record
external buffer info into destructor_arg field.
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
__alloc_skb() cleanup by
Jeff Dike <[email protected]>
include/linux/skbuff.h | 7 ++++---
net/core/skbuff.c | 43 +++++++++++++++++++++++++++++++++++++------
2 files changed, 41 insertions(+), 9 deletions(-)
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 96799f5..8949b15 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -448,17 +448,18 @@ extern void kfree_skb(struct sk_buff *skb);
extern void consume_skb(struct sk_buff *skb);
extern void __kfree_skb(struct sk_buff *skb);
extern struct sk_buff *__alloc_skb(unsigned int size,
- gfp_t priority, int fclone, int node);
+ gfp_t priority, int fclone,
+ int node, struct net_device *dev);
static inline struct sk_buff *alloc_skb(unsigned int size,
gfp_t priority)
{
- return __alloc_skb(size, priority, 0, -1);
+ return __alloc_skb(size, priority, 0, -1, NULL);
}
static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
gfp_t priority)
{
- return __alloc_skb(size, priority, 1, -1);
+ return __alloc_skb(size, priority, 1, -1, NULL);
}
extern int skb_recycle_check(struct sk_buff *skb, int skb_size);
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 6345acc..ae223d2 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -161,7 +161,8 @@ EXPORT_SYMBOL(skb_under_panic);
* @fclone: allocate from fclone cache instead of head cache
* and allocate a cloned (child) skb
* @node: numa node to allocate memory on
- *
+ * @dev: a device owns the skb if the skb try to get external buffer.
+ * otherwise is NULL.
* Allocate a new &sk_buff. The returned buffer has no headroom and a
* tail room of size bytes. The object has a reference count of one.
* The return is the buffer. On a failure the return is %NULL.
@@ -170,12 +171,13 @@ EXPORT_SYMBOL(skb_under_panic);
* %GFP_ATOMIC.
*/
struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
- int fclone, int node)
+ int fclone, int node, struct net_device *dev)
{
struct kmem_cache *cache;
struct skb_shared_info *shinfo;
struct sk_buff *skb;
- u8 *data;
+ u8 *data = NULL;
+ struct skb_external_page *ext_page = NULL;
cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
@@ -185,8 +187,23 @@ struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
goto out;
size = SKB_DATA_ALIGN(size);
- data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
- gfp_mask, node);
+
+ /* If the device wants to do mediate passthru(zero-copy),
+ * the skb may try to get external buffers from outside.
+ * If fails, then fall back to alloc buffers from kernel.
+ */
+ if (dev && dev->mp_port) {
+ ext_page = netdev_alloc_external_page(dev, skb, size);
+ if (ext_page) {
+ data = ext_page->start;
+ size = ext_page->size;
+ }
+ }
+
+ if (!data)
+ data = kmalloc_node_track_caller(
+ size + sizeof(struct skb_shared_info),
+ gfp_mask, node);
if (!data)
goto nodata;
@@ -208,6 +225,15 @@ struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
skb->mac_header = ~0U;
#endif
+ /* If the skb get external buffers sucessfully, since the shinfo is
+ * at the end of the buffer, we may retain the shinfo once we
+ * need it sometime.
+ */
+ if (ext_page) {
+ skb->head = skb->data - NET_IP_ALIGN - NET_SKB_PAD;
+ memcpy(ext_page->ushinfo, skb_shinfo(skb),
+ sizeof(struct skb_shared_info));
+ }
/* make sure we initialize shinfo sequentially */
shinfo = skb_shinfo(skb);
atomic_set(&shinfo->dataref, 1);
@@ -231,6 +257,11 @@ struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
child->fclone = SKB_FCLONE_UNAVAILABLE;
}
+ /* Record the external buffer info in this field. It's not so good,
+ * but we cannot find another place easily.
+ */
+ shinfo->destructor_arg = ext_page;
+
out:
return skb;
nodata:
@@ -259,7 +290,7 @@ struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
struct sk_buff *skb;
- skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
+ skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node, dev);
if (likely(skb)) {
skb_reserve(skb, NET_SKB_PAD);
skb->dev = dev;
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
Export 2 func for device to assign/deassign new strucure
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
include/linux/netdevice.h | 3 +++
net/core/dev.c | 28 ++++++++++++++++++++++++++++
2 files changed, 31 insertions(+), 0 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index bf79756..5c473fb 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1592,6 +1592,9 @@ 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_attach(struct net_device *dev,
+ struct mpassthru_port *port);
+extern void netdev_mp_port_detach(struct net_device *dev);
static inline void napi_free_frags(struct napi_struct *napi)
{
diff --git a/net/core/dev.c b/net/core/dev.c
index e5972f7..6a73fc7 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2464,6 +2464,34 @@ void netif_nit_deliver(struct sk_buff *skb)
rcu_read_unlock();
}
+/* Export two functions to assign/de-assign mp_port pointer
+ * to a net device.
+ */
+
+int netdev_mp_port_attach(struct net_device *dev,
+ struct mpassthru_port *port)
+{
+ /* locked by mp_mutex */
+ if (rcu_dereference(dev->mp_port))
+ return -EBUSY;
+
+ rcu_assign_pointer(dev->mp_port, port);
+
+ return 0;
+}
+EXPORT_SYMBOL(netdev_mp_port_attach);
+
+void netdev_mp_port_detach(struct net_device *dev)
+{
+ /* locked by mp_mutex */
+ if (!rcu_dereference(dev->mp_port))
+ return;
+
+ rcu_assign_pointer(dev->mp_port, NULL);
+ synchronize_rcu();
+}
+EXPORT_SYMBOL(netdev_mp_port_detach);
+
/**
* netif_receive_skb - process receive buffer from network
* @skb: buffer to process
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
If buffer is external, then use the callback to destruct
buffers.
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
net/core/skbuff.c | 11 +++++++++++
1 files changed, 11 insertions(+), 0 deletions(-)
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 169f22c..5d93b2d 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -385,6 +385,11 @@ static void skb_clone_fraglist(struct sk_buff *skb)
static void skb_release_data(struct sk_buff *skb)
{
+ /* check if the skb has external buffers, we have use destructor_arg
+ * here to indicate
+ */
+ struct skb_external_page *ext_page = skb_shinfo(skb)->destructor_arg;
+
if (!skb->cloned ||
!atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
&skb_shinfo(skb)->dataref)) {
@@ -397,6 +402,12 @@ static void skb_release_data(struct sk_buff *skb)
if (skb_has_frags(skb))
skb_drop_fraglist(skb);
+ /* if the skb has external buffers, use destructor here,
+ * since after that skb->head will be kfree, in case skb->head
+ * from external buffer cannot use kfree to destroy.
+ */
+ if (dev_is_mpassthru(skb->dev) && ext_page && ext_page->dtor)
+ ext_page->dtor(ext_page);
kfree(skb->head);
}
}
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
The hook is called in netif_receive_skb().
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
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 4972bc4..3e3ad23 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2543,6 +2543,37 @@ 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
@@ -2622,6 +2653,10 @@ 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
From: Xin Xiaohui <[email protected]>
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
include/linux/netdevice.h | 7 +++++++
1 files changed, 7 insertions(+), 0 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 2f9a4f2..a1a2aaf 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1602,6 +1602,13 @@ extern void netdev_mp_port_detach(struct net_device *dev);
int netdev_mp_port_prep(struct net_device *dev,
struct mpassthru_port *port);
+static int dev_is_mpassthru(struct net_device *dev)
+{
+ if (dev && dev->mp_port)
+ return 1;
+ return 0;
+}
+
static inline void napi_free_frags(struct napi_struct *napi)
{
kfree_skb(napi->skb);
--
1.5.4.4
From: Xin Xiaohui <[email protected]>
The external buffer owner can use the functions to get
the capability of the underlying NIC driver.
Signed-off-by: Xin Xiaohui <[email protected]>
Signed-off-by: Zhao Yu <[email protected]>
Reviewed-by: Jeff Dike <[email protected]>
---
include/linux/netdevice.h | 2 +
net/core/dev.c | 51 +++++++++++++++++++++++++++++++++++++++++++++
2 files changed, 53 insertions(+), 0 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 3a1583b..2f9a4f2 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1599,6 +1599,8 @@ extern gro_result_t napi_gro_frags(struct napi_struct *napi);
extern int netdev_mp_port_attach(struct net_device *dev,
struct mpassthru_port *port);
extern void netdev_mp_port_detach(struct net_device *dev);
+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 6a73fc7..4972bc4 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2492,6 +2492,57 @@ void netdev_mp_port_detach(struct net_device *dev)
}
EXPORT_SYMBOL(netdev_mp_port_detach);
+/* 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;
+
+ /* needed by packet split */
+
+ 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
On Sun, Apr 25, 2010 at 5:19 PM, <[email protected]> wrote:
> +static int dev_is_mpassthru(struct net_device *dev)
> +{
> + if (dev && dev->mp_port)
> + return 1;
> + return 0;
> +}
> +
Please make it a inline function. And you would write it with less
lines of code.
return dev && dev->mp_port;
--
Regards，
Changli Gao([email protected])
On Sun, Apr 25, 2010 at 5:19 PM, <[email protected]> wrote:
> +static int dev_is_mpassthru(struct net_device *dev)
bool return value should be better here.
--
Regards，
Changli Gao([email protected])
From: Changli Gao <[email protected]>
Date: Sun, 25 Apr 2010 17:33:02 +0800
> On Sun, Apr 25, 2010 at 5:19 PM, <[email protected]> wrote:
>> +static int dev_is_mpassthru(struct net_device *dev)
>> +{
>> + ? ? ? if (dev && dev->mp_port)
>> + ? ? ? ? ? ? ? return 1;
>> + ? ? ? return 0;
>> +}
>> +
>
> Please make it a inline function. And you would write it with less
> lines of code.
>
> return dev && dev->mp_port;
And use "bool" :-)
From: Changli Gao <[email protected]>
Date: Sun, 25 Apr 2010 17:35:01 +0800
> On Sun, Apr 25, 2010 at 5:19 PM, <[email protected]> wrote:
>> +static int dev_is_mpassthru(struct net_device *dev)
>
> bool return value should be better here.
Right.
From: [email protected]
Date: Sun, 25 Apr 2010 17:20:06 +0800
> 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.
Isn't it much easier to map the RX ring of the network device into the
guest's address space, have DMA map calls translate guest addresses to
physical/DMA addresses as well as do all of this crazy page pinning
stuff, and provide the translations and protections via the IOMMU?
What's being proposed here looks a bit over-engineered.
On Sun, Apr 25, 2010 at 02:55:29AM -0700, David Miller wrote:
> From: [email protected]
> Date: Sun, 25 Apr 2010 17:20:06 +0800
>
> > 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.
>
> Isn't it much easier to map the RX ring of the network device into the
> guest's address space, have DMA map calls translate guest addresses to
> physical/DMA addresses as well as do all of this crazy page pinning
> stuff, and provide the translations and protections via the IOMMU?
This means we need guest know how the specific network device works.
So we won't be able to, for example, move guest between different hosts.
There are other problems: many physical systems do not have an iommu,
some guest OS-es do not support DMA map calls, doing VM exit
on each DMA map call might turn out to be very slow. And so on.
> What's being proposed here looks a bit over-engineered.
This is an attempt to reduce overhead for virtio (paravirtualization).
'Don't use PV' is kind of an alternative, but I do not
think it's a simpler one.
--
MST
On Sun, Apr 25, 2010 at 05:20:06PM +0800, [email protected] wrote:
> 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 01-12: net core changes.
> patch 13-17: new device as interface to mantpulate external buffers.
> patch 18: for vhost-net.
>
> 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.
>
> Here, we have ever considered 2 ways to utilize the page constructor
> API to dispense the user buffers.
>
> One: Modify __alloc_skb() function a bit, it can only allocate a
> structure of sk_buff, and the data pointer is pointing to a
> user buffer which is coming from a page constructor API.
> Then the shinfo of the skb is also from guest.
> When packet is received from hardware, the skb->data is filled
> directly by h/w. What we have done is in this way.
>
> Pros: We can avoid any copy here.
> Cons: Guest virtio-net driver needs to allocate skb as almost
> the same method with the host NIC drivers, say the size
> of netdev_alloc_skb() and the same reserved space in the
> head of skb. Many NIC drivers are the same with guest and
> ok for this. But some lastest NIC drivers reserves special
> room in skb head. To deal with it, we suggest to provide
> a method in guest virtio-net driver to ask for parameter
> we interest from the NIC driver when we know which device
> we have bind to do zero-copy. Then we ask guest to do so.
> Is that reasonable?
Do you still do this?
> Two: Modify driver to get user buffer allocated from a page constructor
> API(to substitute alloc_page()), the user buffer are used as payload
> buffers and filled by h/w directly when packet is received. Driver
> should associate the pages with skb (skb_shinfo(skb)->frags). For
> the head buffer side, let host allocates skb, and h/w fills it.
> After that, the data filled in host skb header will be copied into
> guest header buffer which is submitted together with the payload buffer.
>
> Pros: We could less care the way how guest or host allocates their
> buffers.
> Cons: We still need a bit copy here for the skb header.
>
> We are not sure which way is the better here. This is the first thing we want
> to get comments from the community. We wish the modification to the network
> part will be generic which not used by vhost-net backend only, but a user
> application may use it as well when the zero-copy device may provides async
> read/write operations later.
I commented on this in the past. Do you still want comments?
> Please give comments especially for the network part modifications.
>
>
> 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. But for simple
> test with netperf, we found bindwidth up and CPU % up too,
> but the bindwidth up ratio is much more than CPU % up ratio.
>
> What we have not done yet:
> packet split support
> To support GRO
> 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.
>
>
> performance:
> using netperf with GSO/TSO disabled, 10G NIC,
> disabled packet split mode, with raw socket case compared to vhost.
>
> bindwidth will be from 1.1Gbps to 1.7Gbps
> CPU % from 120%-140% to 140%-160%
That's nice. The thing to do is probably to enable GSO/TSO
and see what we get this way. Also, mergeable buffer support
was recently posted and I hope to merge it for 2.6.35.
You might want to take a look.
--
MST
On Sun, Apr 25, 2010 at 4:19 AM, <[email protected]> wrote:
> From: Xin Xiaohui <[email protected]>
>
> Signed-off-by: Xin Xiaohui <[email protected]>
> Signed-off-by: Zhao Yu <[email protected]>
> Reviewed-by: Jeff Dike <[email protected]>
> ---
> ?include/linux/netdevice.h | ? 19 ++++++++++++++++++-
> ?1 files changed, 18 insertions(+), 1 deletions(-)
>
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index c79a88b..bf79756 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -530,6 +530,22 @@ 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;
> + ? ? ? struct skb_external_page *(*ctor)(struct mpassthru_port *,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? struct sk_buff *, int);
> +};
I tried searching around, but couldn't find where struct
skb_external_page is declared. Where is it?
Andy
> > 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.
> >
> >Isn't it much easier to map the RX ring of the network device into the
> >guest's address space, have DMA map calls translate guest addresses to
> >physical/DMA addresses as well as do all of this crazy page pinning
> >stuff, and provide the translations and protections via the IOMMU?
>This means we need guest know how the specific network device works.
>So we won't be able to, for example, move guest between different hosts.
>There are other problems: many physical systems do not have an iommu,
>some guest OS-es do not support DMA map calls, doing VM exit
>on each DMA map call might turn out to be very slow. And so on.
This solution is what now we can think of to implement zero-copy. Some
modifications are made to net core to try to avoid network device driver
changes. The major change is to __alloc_skb(), in which we added a dev
parameter to indicate whether the device will DMA to/from guest/user buffer
which is pointed by host skb->data. We also modify skb_release_data() and
skb_reserve(). We made it now works with ixgbe driver with PS mode disabled,
and got some performance data with it.
using netperf with GSO/TSO disabled, 10G NIC,
disabled packet split mode, with raw socket case compared to vhost.
bindwidth will be from 1.1Gbps to 1.7Gbps
CPU % from 120%-140% to 140%-160%
We are now trying to get decent performance data with advanced features.
Do you have any other concerns with this solution?
>> What's being proposed here looks a bit over-engineered.
>This is an attempt to reduce overhead for virtio (paravirtualization).
>'Don't use PV' is kind of an alternative, but I do not
>think it's a simpler one.
--
MST
> +static int dev_is_mpassthru(struct net_device *dev)
>
>bool return value should be better here.
>
>--
>Regards$B!$(B
>Changli Gao([email protected])
Thanks, would fix that.
Thanks
Xiaohui