thr3ads.net - Linux Virtualization - [PATCH net-next V2 11/11] vhost_net: batch submitting XDP buffers to underlayer sockets [Sep 2018]

If this information is useful, please help other people find it:
Share via:

Jason Wang

2018-Sep-12 03:16 UTC

[PATCH net-next V2 00/11] vhost_net TX batching

Hi all:

This series tries to batch submitting packets to underlayer socket
through msg_control during sendmsg(). This is done by:

1) Doing userspace copy inside vhost_net
2) Build XDP buff
3) Batch at most 64 (VHOST_NET_BATCH) XDP buffs and submit them once
   through msg_control during sendmsg().
4) Underlayer sockets can use XDP buffs directly when XDP is enalbed,
   or build skb based on XDP buff.

For the packet that can not be built easily with XDP or for the case
that batch submission is hard (e.g sndbuf is limited). We will go for
the previous slow path, passing iov iterator to underlayer socket
through sendmsg() once per packet.

This can help to improve cache utilization and avoid lots of indirect
calls with sendmsg(). It can also co-operate with the batching support
of the underlayer sockets (e.g the case of XDP redirection through
maps).

Testpmd(txonly) in guest shows obvious improvements:

Test                /+pps%
XDP_DROP on TAP     /+44.8%
XDP_REDIRECT on TAP /+29%
macvtap (skb)       /+26%

Netperf TCP_STREAM TX from guest shows obvious improvements on small
packet:

    size/session/+thu%/+normalize%
       64/     1/   +2%/    0%
       64/     2/   +3%/   +1%
       64/     4/   +7%/   +5%
       64/     8/   +8%/   +6%
      256/     1/   +3%/    0%
      256/     2/  +10%/   +7%
      256/     4/  +26%/  +22%
      256/     8/  +27%/  +23%
      512/     1/   +3%/   +2%
      512/     2/  +19%/  +14%
      512/     4/  +43%/  +40%
      512/     8/  +45%/  +41%
     1024/     1/   +4%/    0%
     1024/     2/  +27%/  +21%
     1024/     4/  +38%/  +73%
     1024/     8/  +15%/  +24%
     2048/     1/  +10%/   +7%
     2048/     2/  +16%/  +12%
     2048/     4/    0%/   +2%
     2048/     8/    0%/   +2%
     4096/     1/  +36%/  +60%
     4096/     2/  -11%/  -26%
     4096/     4/    0%/  +14%
     4096/     8/    0%/   +4%
    16384/     1/   -1%/   +5%
    16384/     2/    0%/   +2%
    16384/     4/    0%/   -3%
    16384/     8/    0%/   +4%
    65535/     1/    0%/  +10%
    65535/     2/    0%/   +8%
    65535/     4/    0%/   +1%
    65535/     8/    0%/   +3%

Please review.

Thanks

Changes from V1:

- don't hold page refcnt for XDP_DROP
- release page if build_skb() fails
- introduce a helper to build skb for the path of no batching mode
- rename tun_do_xdp() as tun_xdp_act() and use negative for reporting
  errors instead of a pointer to integer
- introduce a num field in tun_msg_ctl
- introduce a new structure for storing metadata in the head of XDP
  packet
- do not store batche XDP inside vhoet_net_virtqueue

Jason Wang (11):
  net: sock: introduce SOCK_XDP
  tuntap: switch to use XDP_PACKET_HEADROOM
  tuntap: enable bh early during processing XDP
  tuntap: simplify error handling in tun_build_skb()
  tuntap: tweak on the path of skb XDP case in tun_build_skb()
  tuntap: split out XDP logic
  tuntap: move XDP flushing out of tun_do_xdp()
  tun: switch to new type of msg_control
  tuntap: accept an array of XDP buffs through sendmsg()
  tap: accept an array of XDP buffs through sendmsg()
  vhost_net: batch submitting XDP buffers to underlayer sockets

 drivers/net/tap.c      |  88 +++++++++++++-
 drivers/net/tun.c      | 267 ++++++++++++++++++++++++++++++++---------
 drivers/vhost/net.c    | 181 +++++++++++++++++++++++++---
 include/linux/if_tun.h |  14 +++
 include/net/sock.h     |   1 +
 5 files changed, 471 insertions(+), 80 deletions(-)

-- 
2.17.1

Jason Wang

2018-Sep-12 03:16 UTC

head link

[PATCH net-next V2 01/11] net: sock: introduce SOCK_XDP

This patch introduces a new sock flag - SOCK_XDP. This will be used
for notifying the upper layer that XDP program is attached on the
lower socket, and requires for extra headroom.

TUN will be the first user.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tun.c  | 19 +++++++++++++++++++
 include/net/sock.h |  1 +
 2 files changed, 20 insertions(+)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index ebd07ad82431..2c548bd20393 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -869,6 +869,9 @@ static int tun_attach(struct tun_struct *tun, struct file
*file,
 		tun_napi_init(tun, tfile, napi);
 	}
 
+	if (rtnl_dereference(tun->xdp_prog))
+		sock_set_flag(&tfile->sk, SOCK_XDP);
+
 	tun_set_real_num_queues(tun);
 
 	/* device is allowed to go away first, so no need to hold extra
@@ -1241,13 +1244,29 @@ static int tun_xdp_set(struct net_device *dev, struct
bpf_prog *prog,
 		       struct netlink_ext_ack *extack)
 {
 	struct tun_struct *tun = netdev_priv(dev);
+	struct tun_file *tfile;
 	struct bpf_prog *old_prog;
+	int i;
 
 	old_prog = rtnl_dereference(tun->xdp_prog);
 	rcu_assign_pointer(tun->xdp_prog, prog);
 	if (old_prog)
 		bpf_prog_put(old_prog);
 
+	for (i = 0; i < tun->numqueues; i++) {
+		tfile = rtnl_dereference(tun->tfiles[i]);
+		if (prog)
+			sock_set_flag(&tfile->sk, SOCK_XDP);
+		else
+			sock_reset_flag(&tfile->sk, SOCK_XDP);
+	}
+	list_for_each_entry(tfile, &tun->disabled, next) {
+		if (prog)
+			sock_set_flag(&tfile->sk, SOCK_XDP);
+		else
+			sock_reset_flag(&tfile->sk, SOCK_XDP);
+	}
+
 	return 0;
 }
 
diff --git a/include/net/sock.h b/include/net/sock.h
index 433f45fc2d68..38cae35f6e16 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -800,6 +800,7 @@ enum sock_flags {
 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
 	SOCK_TXTIME,
+	SOCK_XDP, /* XDP is attached */
 };
 
 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL <<
SOCK_TIMESTAMPING_RX_SOFTWARE))
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 02/11] tuntap: switch to use XDP_PACKET_HEADROOM

Acked-by: Michael S. Tsirkin <mst at redhat.com>
Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tun.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 2c548bd20393..d3677a544b56 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -113,7 +113,6 @@ do {								\
 } while (0)
 #endif
 
-#define TUN_HEADROOM 256
 #define TUN_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
 
 /* TUN device flags */
@@ -1654,7 +1653,7 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 	rcu_read_lock();
 	xdp_prog = rcu_dereference(tun->xdp_prog);
 	if (xdp_prog)
-		pad += TUN_HEADROOM;
+		pad += XDP_PACKET_HEADROOM;
 	buflen += SKB_DATA_ALIGN(len + pad);
 	rcu_read_unlock();
 
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 03/11] tuntap: enable bh early during processing XDP

This patch move the bh enabling a little bit earlier, this will be
used for factoring out the core XDP logic of tuntap.

Acked-by: Michael S. Tsirkin <mst at redhat.com>
Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tun.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index d3677a544b56..372caf7d67d9 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1726,22 +1726,18 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 			goto err_xdp;
 		}
 	}
+	rcu_read_unlock();
+	local_bh_enable();
 
 	skb = build_skb(buf, buflen);
-	if (!skb) {
-		rcu_read_unlock();
-		local_bh_enable();
+	if (!skb)
 		return ERR_PTR(-ENOMEM);
-	}
 
 	skb_reserve(skb, pad - delta);
 	skb_put(skb, len);
 	get_page(alloc_frag->page);
 	alloc_frag->offset += buflen;
 
-	rcu_read_unlock();
-	local_bh_enable();
-
 	return skb;
 
 err_redirect:
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 04/11] tuntap: simplify error handling in tun_build_skb()

There's no need to duplicate page get logic in each action. So this
patch tries to get page and calculate the offset before processing XDP
actions (except for XDP_DROP), and undo them when meet errors (we
don't care the performance on errors). This will be used for factoring
out XDP logic.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tun.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 372caf7d67d9..257cf7342d54 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1701,17 +1701,13 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 			xdp_do_flush_map();
 			if (err)
 				goto err_redirect;
-			rcu_read_unlock();
-			local_bh_enable();
-			return NULL;
+			goto out;
 		case XDP_TX:
 			get_page(alloc_frag->page);
 			alloc_frag->offset += buflen;
 			if (tun_xdp_tx(tun->dev, &xdp) < 0)
 				goto err_redirect;
-			rcu_read_unlock();
-			local_bh_enable();
-			return NULL;
+			goto out;
 		case XDP_PASS:
 			delta = orig_data - xdp.data;
 			len = xdp.data_end - xdp.data;
@@ -1742,7 +1738,7 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 
 err_redirect:
 	put_page(alloc_frag->page);
-err_xdp:
+out:
 	rcu_read_unlock();
 	local_bh_enable();
 	this_cpu_inc(tun->pcpu_stats->rx_dropped);
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 05/11] tuntap: tweak on the path of skb XDP case in tun_build_skb()

If we're sure not to go native XDP, there's no need for several things
like bh and rcu stuffs. So this patch introduces a helper to build skb
and hold page refcnt. When we found we will go through skb path, build
skb directly.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tun.c | 39 ++++++++++++++++++++++++---------------
 1 file changed, 24 insertions(+), 15 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 257cf7342d54..946c6148ed75 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1635,6 +1635,23 @@ static bool tun_can_build_skb(struct tun_struct *tun,
struct tun_file *tfile,
 	return true;
 }
 
+static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
+				       int buflen, int len, int pad, int delta)
+{
+	struct sk_buff *skb = build_skb(buf, buflen);
+
+	if (!skb)
+		return ERR_PTR(-ENOMEM);
+
+	skb_reserve(skb, pad - delta);
+	skb_put(skb, len);
+
+	get_page(alloc_frag->page);
+	alloc_frag->offset += buflen;
+
+	return skb;
+}
+
 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
 				     struct tun_file *tfile,
 				     struct iov_iter *from,
@@ -1642,7 +1659,6 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 				     int len, int *skb_xdp)
 {
 	struct page_frag *alloc_frag = &current->task_frag;
-	struct sk_buff *skb;
 	struct bpf_prog *xdp_prog;
 	int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 	unsigned int delta = 0;
@@ -1672,10 +1688,12 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 	 * of xdp_prog above, this should be rare and for simplicity
 	 * we do XDP on skb in case the headroom is not enough.
 	 */
-	if (hdr->gso_type || !xdp_prog)
+	if (hdr->gso_type || !xdp_prog) {
 		*skb_xdp = 1;
-	else
-		*skb_xdp = 0;
+		return __tun_build_skb(alloc_frag, buf, buflen, len, pad, delta);
+	}
+
+	*skb_xdp = 0;
 
 	local_bh_disable();
 	rcu_read_lock();
@@ -1719,22 +1737,13 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 			trace_xdp_exception(tun->dev, xdp_prog, act);
 			/* fall through */
 		case XDP_DROP:
-			goto err_xdp;
+			goto out;
 		}
 	}
 	rcu_read_unlock();
 	local_bh_enable();
 
-	skb = build_skb(buf, buflen);
-	if (!skb)
-		return ERR_PTR(-ENOMEM);
-
-	skb_reserve(skb, pad - delta);
-	skb_put(skb, len);
-	get_page(alloc_frag->page);
-	alloc_frag->offset += buflen;
-
-	return skb;
+	return __tun_build_skb(alloc_frag, buf, buflen, len, pad, delta);
 
 err_redirect:
 	put_page(alloc_frag->page);
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 06/11] tuntap: split out XDP logic

This patch split out XDP logic into a single function. This make it to
be reused by XDP batching path in the following patch.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tun.c | 88 +++++++++++++++++++++++++++--------------------
 1 file changed, 51 insertions(+), 37 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 946c6148ed75..14fe94098180 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1636,14 +1636,14 @@ static bool tun_can_build_skb(struct tun_struct *tun,
struct tun_file *tfile,
 }
 
 static struct sk_buff *__tun_build_skb(struct page_frag *alloc_frag, char *buf,
-				       int buflen, int len, int pad, int delta)
+				       int buflen, int len, int pad)
 {
 	struct sk_buff *skb = build_skb(buf, buflen);
 
 	if (!skb)
 		return ERR_PTR(-ENOMEM);
 
-	skb_reserve(skb, pad - delta);
+	skb_reserve(skb, pad);
 	skb_put(skb, len);
 
 	get_page(alloc_frag->page);
@@ -1652,6 +1652,39 @@ static struct sk_buff *__tun_build_skb(struct page_frag
*alloc_frag, char *buf,
 	return skb;
 }
 
+static int tun_xdp_act(struct tun_struct *tun, struct bpf_prog *xdp_prog,
+		       struct xdp_buff *xdp, u32 act)
+{
+	int err;
+
+	switch (act) {
+	case XDP_REDIRECT:
+		err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
+		xdp_do_flush_map();
+		if (err)
+			return err;
+		break;
+	case XDP_TX:
+		err = tun_xdp_tx(tun->dev, xdp);
+		if (err < 0)
+			return err;
+		break;
+	case XDP_PASS:
+		break;
+	default:
+		bpf_warn_invalid_xdp_action(act);
+		/* fall through */
+	case XDP_ABORTED:
+		trace_xdp_exception(tun->dev, xdp_prog, act);
+		/* fall through */
+	case XDP_DROP:
+		this_cpu_inc(tun->pcpu_stats->rx_dropped);
+		break;
+	}
+
+	return act;
+}
+
 static struct sk_buff *tun_build_skb(struct tun_struct *tun,
 				     struct tun_file *tfile,
 				     struct iov_iter *from,
@@ -1661,10 +1694,10 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 	struct page_frag *alloc_frag = &current->task_frag;
 	struct bpf_prog *xdp_prog;
 	int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
-	unsigned int delta = 0;
 	char *buf;
 	size_t copied;
-	int err, pad = TUN_RX_PAD;
+	int pad = TUN_RX_PAD;
+	int err = 0;
 
 	rcu_read_lock();
 	xdp_prog = rcu_dereference(tun->xdp_prog);
@@ -1690,7 +1723,7 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 	 */
 	if (hdr->gso_type || !xdp_prog) {
 		*skb_xdp = 1;
-		return __tun_build_skb(alloc_frag, buf, buflen, len, pad, delta);
+		return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
 	}
 
 	*skb_xdp = 0;
@@ -1698,9 +1731,8 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 	local_bh_disable();
 	rcu_read_lock();
 	xdp_prog = rcu_dereference(tun->xdp_prog);
-	if (xdp_prog && !*skb_xdp) {
+	if (xdp_prog) {
 		struct xdp_buff xdp;
-		void *orig_data;
 		u32 act;
 
 		xdp.data_hard_start = buf;
@@ -1708,49 +1740,31 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 		xdp_set_data_meta_invalid(&xdp);
 		xdp.data_end = xdp.data + len;
 		xdp.rxq = &tfile->xdp_rxq;
-		orig_data = xdp.data;
-		act = bpf_prog_run_xdp(xdp_prog, &xdp);
 
-		switch (act) {
-		case XDP_REDIRECT:
-			get_page(alloc_frag->page);
-			alloc_frag->offset += buflen;
-			err = xdp_do_redirect(tun->dev, &xdp, xdp_prog);
-			xdp_do_flush_map();
-			if (err)
-				goto err_redirect;
-			goto out;
-		case XDP_TX:
+		act = bpf_prog_run_xdp(xdp_prog, &xdp);
+		if (act == XDP_REDIRECT || act == XDP_TX) {
 			get_page(alloc_frag->page);
 			alloc_frag->offset += buflen;
-			if (tun_xdp_tx(tun->dev, &xdp) < 0)
-				goto err_redirect;
-			goto out;
-		case XDP_PASS:
-			delta = orig_data - xdp.data;
-			len = xdp.data_end - xdp.data;
-			break;
-		default:
-			bpf_warn_invalid_xdp_action(act);
-			/* fall through */
-		case XDP_ABORTED:
-			trace_xdp_exception(tun->dev, xdp_prog, act);
-			/* fall through */
-		case XDP_DROP:
-			goto out;
 		}
+		err = tun_xdp_act(tun, xdp_prog, &xdp, act);
+		if (err < 0)
+			goto err_xdp;
+		if (err != XDP_PASS)
+			goto out;
+
+		pad = xdp.data - xdp.data_hard_start;
+		len = xdp.data_end - xdp.data;
 	}
 	rcu_read_unlock();
 	local_bh_enable();
 
-	return __tun_build_skb(alloc_frag, buf, buflen, len, pad, delta);
+	return __tun_build_skb(alloc_frag, buf, buflen, len, pad);
 
-err_redirect:
+err_xdp:
 	put_page(alloc_frag->page);
 out:
 	rcu_read_unlock();
 	local_bh_enable();
-	this_cpu_inc(tun->pcpu_stats->rx_dropped);
 	return NULL;
 }
 
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 07/11] tuntap: move XDP flushing out of tun_do_xdp()

This will allow adding batch flushing on top.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tun.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 14fe94098180..3ae539374f6b 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1660,7 +1660,6 @@ static int tun_xdp_act(struct tun_struct *tun, struct
bpf_prog *xdp_prog,
 	switch (act) {
 	case XDP_REDIRECT:
 		err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
-		xdp_do_flush_map();
 		if (err)
 			return err;
 		break;
@@ -1749,6 +1748,8 @@ static struct sk_buff *tun_build_skb(struct tun_struct
*tun,
 		err = tun_xdp_act(tun, xdp_prog, &xdp, act);
 		if (err < 0)
 			goto err_xdp;
+		if (err == XDP_REDIRECT)
+			xdp_do_flush_map();
 		if (err != XDP_PASS)
 			goto out;
 
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 08/11] tun: switch to new type of msg_control

This patch introduces to a new tun/tap specific msg_control:

#define TUN_MSG_UBUF 1
#define TUN_MSG_PTR  2
struct tun_msg_ctl {
       int type;
       void *ptr;
};

This allows us to pass different kinds of msg_control through
sendmsg(). The first supported type is ubuf (TUN_MSG_UBUF) which will
be used by the existed vhost_net zerocopy code. The second is XDP
buff, which allows vhost_net to pass XDP buff to TUN. This could be
used to implement accepting an array of XDP buffs from vhost_net in
the following patches.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tap.c      | 18 ++++++++++++------
 drivers/net/tun.c      |  6 +++++-
 drivers/vhost/net.c    |  7 +++++--
 include/linux/if_tun.h | 14 ++++++++++++++
 4 files changed, 36 insertions(+), 9 deletions(-)

diff --git a/drivers/net/tap.c b/drivers/net/tap.c
index f0f7cd977667..7996ed7cbf18 100644
--- a/drivers/net/tap.c
+++ b/drivers/net/tap.c
@@ -619,7 +619,7 @@ static inline struct sk_buff *tap_alloc_skb(struct sock *sk,
size_t prepad,
 #define TAP_RESERVE HH_DATA_OFF(ETH_HLEN)
 
 /* Get packet from user space buffer */
-static ssize_t tap_get_user(struct tap_queue *q, struct msghdr *m,
+static ssize_t tap_get_user(struct tap_queue *q, void *msg_control,
 			    struct iov_iter *from, int noblock)
 {
 	int good_linear = SKB_MAX_HEAD(TAP_RESERVE);
@@ -663,7 +663,7 @@ static ssize_t tap_get_user(struct tap_queue *q, struct
msghdr *m,
 	if (unlikely(len < ETH_HLEN))
 		goto err;
 
-	if (m && m->msg_control && sock_flag(&q->sk,
SOCK_ZEROCOPY)) {
+	if (msg_control && sock_flag(&q->sk, SOCK_ZEROCOPY)) {
 		struct iov_iter i;
 
 		copylen = vnet_hdr.hdr_len ?
@@ -724,11 +724,11 @@ static ssize_t tap_get_user(struct tap_queue *q, struct
msghdr *m,
 	tap = rcu_dereference(q->tap);
 	/* copy skb_ubuf_info for callback when skb has no error */
 	if (zerocopy) {
-		skb_shinfo(skb)->destructor_arg = m->msg_control;
+		skb_shinfo(skb)->destructor_arg = msg_control;
 		skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
 		skb_shinfo(skb)->tx_flags |= SKBTX_SHARED_FRAG;
-	} else if (m && m->msg_control) {
-		struct ubuf_info *uarg = m->msg_control;
+	} else if (msg_control) {
+		struct ubuf_info *uarg = msg_control;
 		uarg->callback(uarg, false);
 	}
 
@@ -1150,7 +1150,13 @@ static int tap_sendmsg(struct socket *sock, struct msghdr
*m,
 		       size_t total_len)
 {
 	struct tap_queue *q = container_of(sock, struct tap_queue, sock);
-	return tap_get_user(q, m, &m->msg_iter, m->msg_flags &
MSG_DONTWAIT);
+	struct tun_msg_ctl *ctl = m->msg_control;
+
+	if (ctl && ctl->type != TUN_MSG_UBUF)
+		return -EINVAL;
+
+	return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
+			    m->msg_flags & MSG_DONTWAIT);
 }
 
 static int tap_recvmsg(struct socket *sock, struct msghdr *m,
diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 3ae539374f6b..89779b58c7ca 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -2431,11 +2431,15 @@ static int tun_sendmsg(struct socket *sock, struct
msghdr *m, size_t total_len)
 	int ret;
 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
 	struct tun_struct *tun = tun_get(tfile);
+	struct tun_msg_ctl *ctl = m->msg_control;
 
 	if (!tun)
 		return -EBADFD;
 
-	ret = tun_get_user(tun, tfile, m->msg_control, &m->msg_iter,
+	if (ctl && ctl->type != TUN_MSG_UBUF)
+		return -EINVAL;
+
+	ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
 			   m->msg_flags & MSG_DONTWAIT,
 			   m->msg_flags & MSG_MORE);
 	tun_put(tun);
diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index 4e656f89cb22..fb01ce6d981c 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -620,6 +620,7 @@ static void handle_tx_zerocopy(struct vhost_net *net, struct
socket *sock)
 		.msg_controllen = 0,
 		.msg_flags = MSG_DONTWAIT,
 	};
+	struct tun_msg_ctl ctl;
 	size_t len, total_len = 0;
 	int err;
 	struct vhost_net_ubuf_ref *uninitialized_var(ubufs);
@@ -664,8 +665,10 @@ static void handle_tx_zerocopy(struct vhost_net *net,
struct socket *sock)
 			ubuf->ctx = nvq->ubufs;
 			ubuf->desc = nvq->upend_idx;
 			refcount_set(&ubuf->refcnt, 1);
-			msg.msg_control = ubuf;
-			msg.msg_controllen = sizeof(ubuf);
+			msg.msg_control = &ctl;
+			ctl.type = TUN_MSG_UBUF;
+			ctl.ptr = ubuf;
+			msg.msg_controllen = sizeof(ctl);
 			ubufs = nvq->ubufs;
 			atomic_inc(&ubufs->refcount);
 			nvq->upend_idx = (nvq->upend_idx + 1) % UIO_MAXIOV;
diff --git a/include/linux/if_tun.h b/include/linux/if_tun.h
index 3d2996dc7d85..12e3eebf0ce6 100644
--- a/include/linux/if_tun.h
+++ b/include/linux/if_tun.h
@@ -16,9 +16,23 @@
 #define __IF_TUN_H
 
 #include <uapi/linux/if_tun.h>
+#include <uapi/linux/virtio_net.h>
 
 #define TUN_XDP_FLAG 0x1UL
 
+#define TUN_MSG_UBUF 1
+#define TUN_MSG_PTR  2
+struct tun_msg_ctl {
+	unsigned short type;
+	unsigned short num;
+	void *ptr;
+};
+
+struct tun_xdp_hdr {
+	int buflen;
+	struct virtio_net_hdr gso;
+};
+
 #if defined(CONFIG_TUN) || defined(CONFIG_TUN_MODULE)
 struct socket *tun_get_socket(struct file *);
 struct ptr_ring *tun_get_tx_ring(struct file *file);
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 09/11] tuntap: accept an array of XDP buffs through sendmsg()

This patch implement TUN_MSG_PTR msg_control type. This type allows
the caller to pass an array of XDP buffs to tuntap through ptr field
of the tun_msg_control. If an XDP program is attached, tuntap can run
XDP program directly. If not, tuntap will build skb and do a fast
receiving since part of the work has been done by vhost_net.

This will avoid lots of indirect calls thus improves the icache
utilization and allows to do XDP batched flushing when doing XDP
redirection.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tun.c | 117 ++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 114 insertions(+), 3 deletions(-)

diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index 89779b58c7ca..2a2cd35853b7 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -2426,22 +2426,133 @@ static void tun_sock_write_space(struct sock *sk)
 	kill_fasync(&tfile->fasync, SIGIO, POLL_OUT);
 }
 
+static int tun_xdp_one(struct tun_struct *tun,
+		       struct tun_file *tfile,
+		       struct xdp_buff *xdp, int *flush)
+{
+	struct tun_xdp_hdr *hdr = xdp->data_hard_start;
+	struct virtio_net_hdr *gso = &hdr->gso;
+	struct tun_pcpu_stats *stats;
+	struct bpf_prog *xdp_prog;
+	struct sk_buff *skb = NULL;
+	u32 rxhash = 0, act;
+	int buflen = hdr->buflen;
+	int err = 0;
+	bool skb_xdp = false;
+
+	xdp_prog = rcu_dereference(tun->xdp_prog);
+	if (xdp_prog) {
+		if (gso->gso_type) {
+			skb_xdp = true;
+			goto build;
+		}
+		xdp_set_data_meta_invalid(xdp);
+		xdp->rxq = &tfile->xdp_rxq;
+
+		act = bpf_prog_run_xdp(xdp_prog, xdp);
+		err = tun_xdp_act(tun, xdp_prog, xdp, act);
+		if (err < 0) {
+			put_page(virt_to_head_page(xdp->data));
+			return err;
+		}
+
+		switch (err) {
+		case XDP_REDIRECT:
+			*flush = true;
+			/* fall through */
+		case XDP_TX:
+			return 0;
+		case XDP_PASS:
+			break;
+		default:
+			put_page(virt_to_head_page(xdp->data));
+			return 0;
+		}
+	}
+
+build:
+	skb = build_skb(xdp->data_hard_start, buflen);
+	if (!skb) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	skb_reserve(skb, xdp->data - xdp->data_hard_start);
+	skb_put(skb, xdp->data_end - xdp->data);
+
+	if (virtio_net_hdr_to_skb(skb, gso, tun_is_little_endian(tun))) {
+		this_cpu_inc(tun->pcpu_stats->rx_frame_errors);
+		kfree_skb(skb);
+		err = -EINVAL;
+		goto out;
+	}
+
+	skb->protocol = eth_type_trans(skb, tun->dev);
+	skb_reset_network_header(skb);
+	skb_probe_transport_header(skb, 0);
+
+	if (skb_xdp) {
+		err = do_xdp_generic(xdp_prog, skb);
+		if (err != XDP_PASS)
+			goto out;
+	}
+
+	if (!rcu_dereference(tun->steering_prog))
+		rxhash = __skb_get_hash_symmetric(skb);
+
+	netif_receive_skb(skb);
+
+	stats = get_cpu_ptr(tun->pcpu_stats);
+	u64_stats_update_begin(&stats->syncp);
+	stats->rx_packets++;
+	stats->rx_bytes += skb->len;
+	u64_stats_update_end(&stats->syncp);
+	put_cpu_ptr(stats);
+
+	if (rxhash)
+		tun_flow_update(tun, rxhash, tfile);
+
+out:
+	return err;
+}
+
 static int tun_sendmsg(struct socket *sock, struct msghdr *m, size_t total_len)
 {
-	int ret;
+	int ret, i;
 	struct tun_file *tfile = container_of(sock, struct tun_file, socket);
 	struct tun_struct *tun = tun_get(tfile);
 	struct tun_msg_ctl *ctl = m->msg_control;
+	struct xdp_buff *xdp;
 
 	if (!tun)
 		return -EBADFD;
 
-	if (ctl && ctl->type != TUN_MSG_UBUF)
-		return -EINVAL;
+	if (ctl && (ctl->type == TUN_MSG_PTR)) {
+		int n = ctl->num;
+		int flush = 0;
+
+		local_bh_disable();
+		rcu_read_lock();
+
+		for (i = 0; i < n; i++) {
+			xdp = &((struct xdp_buff *)ctl->ptr)[i];
+			tun_xdp_one(tun, tfile, xdp, &flush);
+		}
+
+		if (flush)
+			xdp_do_flush_map();
+
+		rcu_read_unlock();
+		local_bh_enable();
+
+		ret = total_len;
+		goto out;
+	}
 
 	ret = tun_get_user(tun, tfile, ctl ? ctl->ptr : NULL, &m->msg_iter,
 			   m->msg_flags & MSG_DONTWAIT,
 			   m->msg_flags & MSG_MORE);
+out:
 	tun_put(tun);
 	return ret;
 }
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 10/11] tap: accept an array of XDP buffs through sendmsg()

This patch implement TUN_MSG_PTR msg_control type. This type allows
the caller to pass an array of XDP buffs to tuntap through ptr field
of the tun_msg_control. Tap will build skb through those XDP buffers.

This will avoid lots of indirect calls thus improves the icache
utilization and allows to do XDP batched flushing when doing XDP
redirection.

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/net/tap.c | 74 +++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 72 insertions(+), 2 deletions(-)

diff --git a/drivers/net/tap.c b/drivers/net/tap.c
index 7996ed7cbf18..a4ab4a791fe7 100644
--- a/drivers/net/tap.c
+++ b/drivers/net/tap.c
@@ -1146,14 +1146,84 @@ static const struct file_operations tap_fops = {
 #endif
 };
 
+static int tap_get_user_xdp(struct tap_queue *q, struct xdp_buff *xdp)
+{
+	struct tun_xdp_hdr *hdr = xdp->data_hard_start;
+	struct virtio_net_hdr *gso = &hdr->gso;
+	int buflen = hdr->buflen;
+	int vnet_hdr_len = 0;
+	struct tap_dev *tap;
+	struct sk_buff *skb;
+	int err, depth;
+
+	if (q->flags & IFF_VNET_HDR)
+		vnet_hdr_len = READ_ONCE(q->vnet_hdr_sz);
+
+	skb = build_skb(xdp->data_hard_start, buflen);
+	if (!skb) {
+		err = -ENOMEM;
+		goto err;
+	}
+
+	skb_reserve(skb, xdp->data - xdp->data_hard_start);
+	skb_put(skb, xdp->data_end - xdp->data);
+
+	skb_set_network_header(skb, ETH_HLEN);
+	skb_reset_mac_header(skb);
+	skb->protocol = eth_hdr(skb)->h_proto;
+
+	if (vnet_hdr_len) {
+		err = virtio_net_hdr_to_skb(skb, gso, tap_is_little_endian(q));
+		if (err)
+			goto err_kfree;
+	}
+
+	skb_probe_transport_header(skb, ETH_HLEN);
+
+	/* Move network header to the right position for VLAN tagged packets */
+	if ((skb->protocol == htons(ETH_P_8021Q) ||
+	     skb->protocol == htons(ETH_P_8021AD)) &&
+	    __vlan_get_protocol(skb, skb->protocol, &depth) != 0)
+		skb_set_network_header(skb, depth);
+
+	rcu_read_lock();
+	tap = rcu_dereference(q->tap);
+	if (tap) {
+		skb->dev = tap->dev;
+		dev_queue_xmit(skb);
+	} else {
+		kfree_skb(skb);
+	}
+	rcu_read_unlock();
+
+	return 0;
+
+err_kfree:
+	kfree_skb(skb);
+err:
+	rcu_read_lock();
+		tap = rcu_dereference(q->tap);
+	if (tap && tap->count_tx_dropped)
+		tap->count_tx_dropped(tap);
+	rcu_read_unlock();
+	return err;
+}
+
 static int tap_sendmsg(struct socket *sock, struct msghdr *m,
 		       size_t total_len)
 {
 	struct tap_queue *q = container_of(sock, struct tap_queue, sock);
 	struct tun_msg_ctl *ctl = m->msg_control;
+	struct xdp_buff *xdp;
+	int i;
 
-	if (ctl && ctl->type != TUN_MSG_UBUF)
-		return -EINVAL;
+	if (ctl && (ctl->type == TUN_MSG_PTR)) {
+		for (i = 0; i < ctl->num; i++) {
+			xdp = &((struct xdp_buff *)ctl->ptr)[i];
+			tap_get_user_xdp(q, xdp);
+		}
+		return 0;
+	}
 
 	return tap_get_user(q, ctl ? ctl->ptr : NULL, &m->msg_iter,
 			    m->msg_flags & MSG_DONTWAIT);
-- 
2.17.1

Jason Wang

2018-Sep-12 03:17 UTC

head link

[PATCH net-next V2 11/11] vhost_net: batch submitting XDP buffers to underlayer sockets

This patch implements XDP batching for vhost_net. The idea is first to
try to do userspace copy and build XDP buff directly in vhost. Instead
of submitting the packet immediately, vhost_net will batch them in an
array and submit every 64 (VHOST_NET_BATCH) packets to the under layer
sockets through msg_control of sendmsg().

When XDP is enabled on the TUN/TAP, TUN/TAP can process XDP inside a
loop without caring GUP thus it can do batch map flushing. When XDP is
not enabled or not supported, the underlayer socket need to build skb
and pass it to network core. The batched packet submission allows us
to do batching like netif_receive_skb_list() in the future.

This saves lots of indirect calls for better cache utilization. For
the case that we can't so batching e.g when sndbuf is limited or
packet size is too large, we will go for usual one packet per
sendmsg() way.

Doing testpmd on various setups gives us:

Test                /+pps%
XDP_DROP on TAP     /+44.8%
XDP_REDIRECT on TAP /+29%
macvtap (skb)       /+26%

Netperf tests shows obvious improvements for small packet transmission:

size/session/+thu%/+normalize%
   64/     1/   +2%/    0%
   64/     2/   +3%/   +1%
   64/     4/   +7%/   +5%
   64/     8/   +8%/   +6%
  256/     1/   +3%/    0%
  256/     2/  +10%/   +7%
  256/     4/  +26%/  +22%
  256/     8/  +27%/  +23%
  512/     1/   +3%/   +2%
  512/     2/  +19%/  +14%
  512/     4/  +43%/  +40%
  512/     8/  +45%/  +41%
 1024/     1/   +4%/    0%
 1024/     2/  +27%/  +21%
 1024/     4/  +38%/  +73%
 1024/     8/  +15%/  +24%
 2048/     1/  +10%/   +7%
 2048/     2/  +16%/  +12%
 2048/     4/    0%/   +2%
 2048/     8/    0%/   +2%
 4096/     1/  +36%/  +60%
 4096/     2/  -11%/  -26%
 4096/     4/    0%/  +14%
 4096/     8/    0%/   +4%
16384/     1/   -1%/   +5%
16384/     2/    0%/   +2%
16384/     4/    0%/   -3%
16384/     8/    0%/   +4%
65535/     1/    0%/  +10%
65535/     2/    0%/   +8%
65535/     4/    0%/   +1%
65535/     8/    0%/   +3%

Signed-off-by: Jason Wang <jasowang at redhat.com>
---
 drivers/vhost/net.c | 174 ++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 161 insertions(+), 13 deletions(-)

diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index fb01ce6d981c..dd4e0a301635 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -116,6 +116,8 @@ struct vhost_net_virtqueue {
 	 * For RX, number of batched heads
 	 */
 	int done_idx;
+	/* Number of XDP frames batched */
+	int batched_xdp;
 	/* an array of userspace buffers info */
 	struct ubuf_info *ubuf_info;
 	/* Reference counting for outstanding ubufs.
@@ -123,6 +125,8 @@ struct vhost_net_virtqueue {
 	struct vhost_net_ubuf_ref *ubufs;
 	struct ptr_ring *rx_ring;
 	struct vhost_net_buf rxq;
+	/* Batched XDP buffs */
+	struct xdp_buff *xdp;
 };
 
 struct vhost_net {
@@ -338,6 +342,11 @@ static bool vhost_sock_zcopy(struct socket *sock)
 		sock_flag(sock->sk, SOCK_ZEROCOPY);
 }
 
+static bool vhost_sock_xdp(struct socket *sock)
+{
+	return sock_flag(sock->sk, SOCK_XDP);
+}
+
 /* In case of DMA done not in order in lower device driver for some reason.
  * upend_idx is used to track end of used idx, done_idx is used to track head
  * of used idx. Once lower device DMA done contiguously, we will signal KVM
@@ -444,10 +453,37 @@ static void vhost_net_signal_used(struct
vhost_net_virtqueue *nvq)
 	nvq->done_idx = 0;
 }
 
+static void vhost_tx_batch(struct vhost_net *net,
+			   struct vhost_net_virtqueue *nvq,
+			   struct socket *sock,
+			   struct msghdr *msghdr)
+{
+	struct tun_msg_ctl ctl = {
+		.type = TUN_MSG_PTR,
+		.num = nvq->batched_xdp,
+		.ptr = nvq->xdp,
+	};
+	int err;
+
+	if (nvq->batched_xdp == 0)
+		goto signal_used;
+
+	msghdr->msg_control = &ctl;
+	err = sock->ops->sendmsg(sock, msghdr, 0);
+	if (unlikely(err < 0)) {
+		vq_err(&nvq->vq, "Fail to batch sending packets\n");
+		return;
+	}
+
+signal_used:
+	vhost_net_signal_used(nvq);
+	nvq->batched_xdp = 0;
+}
+
 static int vhost_net_tx_get_vq_desc(struct vhost_net *net,
 				    struct vhost_net_virtqueue *nvq,
 				    unsigned int *out_num, unsigned int *in_num,
-				    bool *busyloop_intr)
+				    struct msghdr *msghdr, bool *busyloop_intr)
 {
 	struct vhost_virtqueue *vq = &nvq->vq;
 	unsigned long uninitialized_var(endtime);
@@ -455,8 +491,9 @@ static int vhost_net_tx_get_vq_desc(struct vhost_net *net,
 				  out_num, in_num, NULL, NULL);
 
 	if (r == vq->num && vq->busyloop_timeout) {
+		/* Flush batched packets first */
 		if (!vhost_sock_zcopy(vq->private_data))
-			vhost_net_signal_used(nvq);
+			vhost_tx_batch(net, nvq, vq->private_data, msghdr);
 		preempt_disable();
 		endtime = busy_clock() + vq->busyloop_timeout;
 		while (vhost_can_busy_poll(endtime)) {
@@ -512,7 +549,7 @@ static int get_tx_bufs(struct vhost_net *net,
 	struct vhost_virtqueue *vq = &nvq->vq;
 	int ret;
 
-	ret = vhost_net_tx_get_vq_desc(net, nvq, out, in, busyloop_intr);
+	ret = vhost_net_tx_get_vq_desc(net, nvq, out, in, msg, busyloop_intr);
 
 	if (ret < 0 || ret == vq->num)
 		return ret;
@@ -540,6 +577,80 @@ static bool tx_can_batch(struct vhost_virtqueue *vq, size_t
total_len)
 	       !vhost_vq_avail_empty(vq->dev, vq);
 }
 
+#define VHOST_NET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
+
+static int vhost_net_build_xdp(struct vhost_net_virtqueue *nvq,
+			       struct iov_iter *from)
+{
+	struct vhost_virtqueue *vq = &nvq->vq;
+	struct socket *sock = vq->private_data;
+	struct page_frag *alloc_frag = &current->task_frag;
+	struct virtio_net_hdr *gso;
+	struct xdp_buff *xdp = &nvq->xdp[nvq->batched_xdp];
+	struct tun_xdp_hdr *hdr;
+	size_t len = iov_iter_count(from);
+	int headroom = vhost_sock_xdp(sock) ? XDP_PACKET_HEADROOM : 0;
+	int buflen = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
+	int pad = SKB_DATA_ALIGN(VHOST_NET_RX_PAD + headroom + nvq->sock_hlen);
+	int sock_hlen = nvq->sock_hlen;
+	void *buf;
+	int copied;
+
+	if (unlikely(len < nvq->sock_hlen))
+		return -EFAULT;
+
+	if (SKB_DATA_ALIGN(len + pad) +
+	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) > PAGE_SIZE)
+		return -ENOSPC;
+
+	buflen += SKB_DATA_ALIGN(len + pad);
+	alloc_frag->offset = ALIGN((u64)alloc_frag->offset, SMP_CACHE_BYTES);
+	if (unlikely(!skb_page_frag_refill(buflen, alloc_frag, GFP_KERNEL)))
+		return -ENOMEM;
+
+	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
+	copied = copy_page_from_iter(alloc_frag->page,
+				     alloc_frag->offset +
+				     offsetof(struct tun_xdp_hdr, gso),
+				     sock_hlen, from);
+	if (copied != sock_hlen)
+		return -EFAULT;
+
+	hdr = buf;
+	gso = &hdr->gso;
+
+	if ((gso->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
+	    vhost16_to_cpu(vq, gso->csum_start) +
+	    vhost16_to_cpu(vq, gso->csum_offset) + 2 >
+	    vhost16_to_cpu(vq, gso->hdr_len)) {
+		gso->hdr_len = cpu_to_vhost16(vq,
+			       vhost16_to_cpu(vq, gso->csum_start) +
+			       vhost16_to_cpu(vq, gso->csum_offset) + 2);
+
+		if (vhost16_to_cpu(vq, gso->hdr_len) > len)
+			return -EINVAL;
+	}
+
+	len -= sock_hlen;
+	copied = copy_page_from_iter(alloc_frag->page,
+				     alloc_frag->offset + pad,
+				     len, from);
+	if (copied != len)
+		return -EFAULT;
+
+	xdp->data_hard_start = buf;
+	xdp->data = buf + pad;
+	xdp->data_end = xdp->data + len;
+	hdr->buflen = buflen;
+
+	get_page(alloc_frag->page);
+	alloc_frag->offset += buflen;
+
+	++nvq->batched_xdp;
+
+	return 0;
+}
+
 static void handle_tx_copy(struct vhost_net *net, struct socket *sock)
 {
 	struct vhost_net_virtqueue *nvq = &net->vqs[VHOST_NET_VQ_TX];
@@ -556,10 +667,14 @@ static void handle_tx_copy(struct vhost_net *net, struct
socket *sock)
 	size_t len, total_len = 0;
 	int err;
 	int sent_pkts = 0;
+	bool sock_can_batch = (sock->sk->sk_sndbuf == INT_MAX);
 
 	for (;;) {
 		bool busyloop_intr = false;
 
+		if (nvq->done_idx == VHOST_NET_BATCH)
+			vhost_tx_batch(net, nvq, sock, &msg);
+
 		head = get_tx_bufs(net, nvq, &msg, &out, &in, &len,
 				   &busyloop_intr);
 		/* On error, stop handling until the next kick. */
@@ -577,14 +692,34 @@ static void handle_tx_copy(struct vhost_net *net, struct
socket *sock)
 			break;
 		}
 
-		vq->heads[nvq->done_idx].id = cpu_to_vhost32(vq, head);
-		vq->heads[nvq->done_idx].len = 0;
-
 		total_len += len;
-		if (tx_can_batch(vq, total_len))
-			msg.msg_flags |= MSG_MORE;
-		else
-			msg.msg_flags &= ~MSG_MORE;
+
+		/* For simplicity, TX batching is only enabled if
+		 * sndbuf is unlimited.
+		 */
+		if (sock_can_batch) {
+			err = vhost_net_build_xdp(nvq, &msg.msg_iter);
+			if (!err) {
+				goto done;
+			} else if (unlikely(err != -ENOSPC)) {
+				vhost_tx_batch(net, nvq, sock, &msg);
+				vhost_discard_vq_desc(vq, 1);
+				vhost_net_enable_vq(net, vq);
+				break;
+			}
+
+			/* We can't build XDP buff, go for single
+			 * packet path but let's flush batched
+			 * packets.
+			 */
+			vhost_tx_batch(net, nvq, sock, &msg);
+			msg.msg_control = NULL;
+		} else {
+			if (tx_can_batch(vq, total_len))
+				msg.msg_flags |= MSG_MORE;
+			else
+				msg.msg_flags &= ~MSG_MORE;
+		}
 
 		/* TODO: Check specific error and bomb out unless ENOBUFS? */
 		err = sock->ops->sendmsg(sock, &msg, len);
@@ -596,15 +731,17 @@ static void handle_tx_copy(struct vhost_net *net, struct
socket *sock)
 		if (err != len)
 			pr_debug("Truncated TX packet: len %d != %zd\n",
 				 err, len);
-		if (++nvq->done_idx >= VHOST_NET_BATCH)
-			vhost_net_signal_used(nvq);
+done:
+		vq->heads[nvq->done_idx].id = cpu_to_vhost32(vq, head);
+		vq->heads[nvq->done_idx].len = 0;
+		++nvq->done_idx;
 		if (vhost_exceeds_weight(++sent_pkts, total_len)) {
 			vhost_poll_queue(&vq->poll);
 			break;
 		}
 	}
 
-	vhost_net_signal_used(nvq);
+	vhost_tx_batch(net, nvq, sock, &msg);
 }
 
 static void handle_tx_zerocopy(struct vhost_net *net, struct socket *sock)
@@ -1081,6 +1218,7 @@ static int vhost_net_open(struct inode *inode, struct file
*f)
 	struct vhost_dev *dev;
 	struct vhost_virtqueue **vqs;
 	void **queue;
+	struct xdp_buff *xdp;
 	int i;
 
 	n = kvmalloc(sizeof *n, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
@@ -1101,6 +1239,14 @@ static int vhost_net_open(struct inode *inode, struct
file *f)
 	}
 	n->vqs[VHOST_NET_VQ_RX].rxq.queue = queue;
 
+	xdp = kmalloc_array(VHOST_NET_BATCH, sizeof(*xdp), GFP_KERNEL);
+	if (!xdp) {
+		kfree(vqs);
+		kvfree(n);
+		kfree(queue);
+	}
+	n->vqs[VHOST_NET_VQ_TX].xdp = xdp;
+
 	dev = &n->dev;
 	vqs[VHOST_NET_VQ_TX] = &n->vqs[VHOST_NET_VQ_TX].vq;
 	vqs[VHOST_NET_VQ_RX] = &n->vqs[VHOST_NET_VQ_RX].vq;
@@ -1111,6 +1257,7 @@ static int vhost_net_open(struct inode *inode, struct file
*f)
 		n->vqs[i].ubuf_info = NULL;
 		n->vqs[i].upend_idx = 0;
 		n->vqs[i].done_idx = 0;
+		n->vqs[i].batched_xdp = 0;
 		n->vqs[i].vhost_hlen = 0;
 		n->vqs[i].sock_hlen = 0;
 		n->vqs[i].rx_ring = NULL;
@@ -1194,6 +1341,7 @@ static int vhost_net_release(struct inode *inode, struct
file *f)
 	 * since jobs can re-queue themselves. */
 	vhost_net_flush(n);
 	kfree(n->vqs[VHOST_NET_VQ_RX].rxq.queue);
+	kfree(n->vqs[VHOST_NET_VQ_TX].xdp);
 	kfree(n->dev.vqs);
 	kvfree(n);
 	return 0;
-- 
2.17.1

David Miller

2018-Sep-13 16:28 UTC

head link

[PATCH net-next V2 00/11] vhost_net TX batching

From: Jason Wang <jasowang at redhat.com>
Date: Wed, 12 Sep 2018 11:16:58 +0800
> This series tries to batch submitting packets to underlayer socket
> through msg_control during sendmsg(). This is done by: ...

Series applied, thanks Jason.

Michael S. Tsirkin

2018-Sep-13 18:02 UTC

head link

[PATCH net-next V2 00/11] vhost_net TX batching

On Thu, Sep 13, 2018 at 09:28:19AM -0700, David Miller
wrote:> From: Jason Wang <jasowang at redhat.com>
> Date: Wed, 12 Sep 2018 11:16:58 +0800
> 
> > This series tries to batch submitting packets to underlayer socket
> > through msg_control during sendmsg(). This is done by:
>  ...
> 
> Series applied, thanks Jason.
Going over it now I don't see a lot to complain about, but I'd
appreciate a bit more time for review in the future. 3 days ok?

-- 
MST

Michael S. Tsirkin

2018-Sep-13 18:04 UTC

head link

[PATCH net-next V2 11/11] vhost_net: batch submitting XDP buffers to underlayer sockets

On Wed, Sep 12, 2018 at 11:17:09AM +0800, Jason Wang
wrote:> +static void vhost_tx_batch(struct vhost_net *net,
> +			   struct vhost_net_virtqueue *nvq,
> +			   struct socket *sock,
> +			   struct msghdr *msghdr)
> +{
> +	struct tun_msg_ctl ctl = {
> +		.type = TUN_MSG_PTR,
> +		.num = nvq->batched_xdp,
> +		.ptr = nvq->xdp,
> +	};
> +	int err;
> +
> +	if (nvq->batched_xdp == 0)
> +		goto signal_used;
> +
> +	msghdr->msg_control = &ctl;
> +	err = sock->ops->sendmsg(sock, msghdr, 0);
> +	if (unlikely(err < 0)) {
> +		vq_err(&nvq->vq, "Fail to batch sending packets\n");
> +		return;
> +	}
> +
> +signal_used:
> +	vhost_net_signal_used(nvq);
> +	nvq->batched_xdp = 0;
> +}
> +
Given it's all tun-specific now, how about just exporting tap_sendmsg
and calling that? Will get rid of some indirection too.

-- 
MST

Reasonably Related Threads

Search for more apparently analagous threads

Linux Virtualization - Sep 2018 - [PATCH net-next V2 11/11] vhost_net: batch submitting XDP buffers to underlayer sockets

[PATCH net-next V2 00/11] vhost_net TX batching

[PATCH net-next V2 01/11] net: sock: introduce SOCK_XDP

[PATCH net-next V2 02/11] tuntap: switch to use XDP_PACKET_HEADROOM

[PATCH net-next V2 03/11] tuntap: enable bh early during processing XDP

[PATCH net-next V2 04/11] tuntap: simplify error handling in tun_build_skb()

[PATCH net-next V2 05/11] tuntap: tweak on the path of skb XDP case in tun_build_skb()

[PATCH net-next V2 06/11] tuntap: split out XDP logic

[PATCH net-next V2 07/11] tuntap: move XDP flushing out of tun_do_xdp()

[PATCH net-next V2 08/11] tun: switch to new type of msg_control

[PATCH net-next V2 09/11] tuntap: accept an array of XDP buffs through sendmsg()

[PATCH net-next V2 10/11] tap: accept an array of XDP buffs through sendmsg()

[PATCH net-next V2 11/11] vhost_net: batch submitting XDP buffers to underlayer sockets

[PATCH net-next V2 00/11] vhost_net TX batching

[PATCH net-next V2 00/11] vhost_net TX batching

[PATCH net-next V2 11/11] vhost_net: batch submitting XDP buffers to underlayer sockets

Reasonably Related Threads