Skip to contentSkip to conversation

netdev: handle setting transmit queue length on active device.

PATCH In review Stephen Hemminger 3 participants 9 messages Mar 18, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 netdev: handle setting transmit queue length on active device.

Some virtual devices like VLAN's or bridges by default have not transmit queue. In order to do queuing disciplines on these devices a queue must be added by setting transmit queue length. The kernel allows doing this at any time, but setting the queue length value is not enough. Setting the queue length does not create the transmit queue (going from 0 to non-zero) or destroy it (setting to zero); the queue is only created (or destroyed) when device comes up (or goes down).

This patch handles this be doing the necessary activations.

---
 include/linux/netdevice.h |    2 ++
 net/core/dev.c            |   19 +++++++++++++++++--
 net/core/net-sysfs.c      |    5 ++---
 net/core/rtnetlink.c      |    2 +-
 4 files changed, 22 insertions(+), 6 deletions(-)
include/linux/netdevice.h +2 -0
@@ -1436,6 +1436,8 @@ extern int		dev_change_net_namespace(str
 extern int		dev_set_mtu(struct net_device *, int);
 extern int		dev_set_mac_address(struct net_device *,
 					    struct sockaddr *);
+extern int		dev_set_tx_queue_len(struct net_device *dev,
+					     unsigned long len);
 extern int		dev_hard_start_xmit(struct sk_buff *skb,
 					    struct net_device *dev,
 					    struct netdev_queue *txq);
net/core/dev.c +17 -2
@@ -983,6 +983,22 @@ int dev_set_alias(struct net_device *dev
 	return len;
 }
 
+/**
+ *
+ *	@dev: device
+ *	@len: limit of frames in transmit queue
+ *
+ *	Set ifalias for a device,
+ */
+int dev_set_tx_queue_len(struct net_device *dev, unsigned long len)
+{
+	if (dev->flags & IFF_UP)
+		dev_deactivate(dev);
+	dev->tx_queue_len = len;
+	if (dev->flags & IFF_UP)
+		dev_activate(dev);
+	return 0;
+}
 
 /**
  *	netdev_features_change - device changes features
@@ -3975,8 +3991,7 @@ static int dev_ifsioc(struct net *net, s
 		case SIOCSIFTXQLEN:
 			if (ifr->ifr_qlen < 0)
 				return -EINVAL;
-			dev->tx_queue_len = ifr->ifr_qlen;
-			return 0;
+			return dev_set_tx_queue_len(dev, ifr->ifr_qlen);
 
 		case SIOCSIFNAME:
 			ifr->ifr_newname[IFNAMSIZ-1] = '\0';
net/core/net-sysfs.c +2 -3
@@ -198,10 +198,9 @@ static ssize_t store_flags(struct device
 
 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 
-static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
+static int change_tx_queue_len(struct net_device *netdev, unsigned long new_len)
 {
-	net->tx_queue_len = new_len;
-	return 0;
+	return dev_set_tx_queue_len(netdev, new_len);
 }
 
 static ssize_t store_tx_queue_len(struct device *dev,
net/core/rtnetlink.c +1 -1
@@ -885,7 +885,7 @@ static int do_setlink(struct net_device 
 	}
 
 	if (tb[IFLA_TXQLEN])
-		dev->tx_queue_len = nla_get_u32(tb[IFLA_TXQLEN]);
+		dev_set_tx_queue_len(dev, nla_get_u32(tb[IFLA_TXQLEN]));
 
 	if (tb[IFLA_OPERSTATE])
 		set_operstate(dev, nla_get_u8(tb[IFLA_OPERSTATE]));
#1 lore raw
Quoted 18 lines · Stephen Hemminger

Some virtual devices like VLAN's or bridges by default have not transmit queue. In order to do queuing disciplines on these devices a queue must be added by setting transmit queue length. The kernel allows doing this at any time, but setting the queue length value is not enough. Setting the queue length does not create the transmit queue (going from 0 to non-zero) or destroy it (setting to zero); the queue is only created (or destroyed) when device comes up (or goes down).

This patch handles this be doing the necessary activations.

net/core/dev.c +6 -0
+/**
+ *
+ *	@dev: device
+ *	@len: limit of frames in transmit queue
+ *
+ *	Set ifalias for a device,

net/core/dev.c int dev_set_alias(struct net_device *dev · line 1001

Quoted 10 lines
net/core/dev.c +10 -0
+ */
+int dev_set_tx_queue_len(struct net_device *dev, unsigned long len)
+{
+	if (dev->flags & IFF_UP)
+		dev_deactivate(dev);
+	dev->tx_queue_len = len;
+	if (dev->flags & IFF_UP)
+		dev_activate(dev);
+	return 0;
+}

That seems quite heavy-weight in case of hierarchical qdiscs, it will walk and reset the entire tree and looses all state, including packets as a side effect. Its also a bit inconsistent since it only affects the default pfifos at the root, not the inner default qdiscs if I'm not mistaken.

A less intrusive way would be to simply use dev->tx_queue_len directly in pfifo when no limit is configured. That would require to make a runtime decision for each packet though. Two more alternatives:

- add a special pfifo variant that always uses tx_queue_len. Duplication comes down to pfifo_enqueue (a few lines) and a new struct Qdisc_ops

- if you really only care about the root, reconfigure it explicitly using fifo_set_limit()

#2 lore raw
Quoted 41 lines · On Wed, 18 Mar 2009 18:15:38 +0100 Patrick McHardy <[email protected]> wrote:
Quoted 18 lines · Stephen Hemminger

Some virtual devices like VLAN's or bridges by default have not transmit queue. In order to do queuing disciplines on these devices a queue must be added by setting transmit queue length. The kernel allows doing this at any time, but setting the queue length value is not enough. Setting the queue length does not create the transmit queue (going from 0 to non-zero) or destroy it (setting to zero); the queue is only created (or destroyed) when device comes up (or goes down).

This patch handles this be doing the necessary activations.

net/core/dev.c +6 -0
+/**
+ *
+ *	@dev: device
+ *	@len: limit of frames in transmit queue
+ *
+ *	Set ifalias for a device,

net/core/dev.c int dev_set_alias(struct net_device *dev · line 1001

Quoted 10 lines
net/core/dev.c +10 -0
+ */
+int dev_set_tx_queue_len(struct net_device *dev, unsigned long len)
+{
+	if (dev->flags & IFF_UP)
+		dev_deactivate(dev);
+	dev->tx_queue_len = len;
+	if (dev->flags & IFF_UP)
+		dev_activate(dev);
+	return 0;
+}

That seems quite heavy-weight in case of hierarchical qdiscs, it will walk and reset the entire tree and looses all state, including packets as a side effect. Its also a bit inconsistent since it only affects the default pfifos at the root, not the inner default qdiscs if I'm not mistaken.

A less intrusive way would be to simply use dev->tx_queue_len directly in pfifo when no limit is configured. That would require to make a runtime decision for each packet though. Two more alternatives:

Quoted 6 lines

- add a special pfifo variant that always uses tx_queue_len. Duplication comes down to pfifo_enqueue (a few lines) and a new struct Qdisc_ops

- if you really only care about the root, reconfigure it explicitly using fifo_set_limit()

The problem is when setting limit to 0 to force noop and setting it to non-zero to force pfifo_fast back again.

#3 lore raw
Quoted 27 lines · Stephen Hemminger
Quoted 24 lines · On Wed, 18 Mar 2009 18:15:38 +0100 Patrick McHardy <[email protected]> wrote:
Quoted 10 lines · Stephen Hemminger

Some virtual devices like VLAN's or bridges by default have not transmit queue. In order to do queuing disciplines on these devices a queue must be added by setting transmit queue length. The kernel allows doing this at any time, but setting the queue length value is not enough. Setting the queue length does not create the transmit queue (going from 0 to non-zero) or destroy it (setting to zero); the queue is only created (or destroyed) when device comes up (or goes down).

This patch handles this be doing the necessary activations.

That seems quite heavy-weight in case of hierarchical qdiscs, it will walk and reset the entire tree and looses all state, including packets as a side effect. Its also a bit inconsistent since it only affects the default pfifos at the root, not the inner default qdiscs if I'm not mistaken.

A less intrusive way would be to simply use dev->tx_queue_len directly in pfifo when no limit is configured. That would require to make a runtime decision for each packet though. Two more alternatives:

- add a special pfifo variant that always uses tx_queue_len. Duplication comes down to pfifo_enqueue (a few lines) and a new struct Qdisc_ops

Quoted 6 lines
Quoted 3 lines

- if you really only care about the root, reconfigure it explicitly using fifo_set_limit()

The problem is when setting limit to 0 to force noop and setting it to non-zero to force pfifo_fast back again.

I'd suggest to simply attach a pfifo qdisc with an explicitly configured limit to activate queueing on virtual devices.

#4 lore raw
Quoted 32 lines · On Wed, 18 Mar 2009 19:08:07 +0100 Patrick McHardy <[email protected]> wrote:
Quoted 31 lines · Stephen Hemminger
Quoted 27 lines · On Wed, 18 Mar 2009 18:15:38 +0100 Patrick McHardy <[email protected]> wrote:
Quoted 10 lines · Stephen Hemminger

Some virtual devices like VLAN's or bridges by default have not transmit queue. In order to do queuing disciplines on these devices a queue must be added by setting transmit queue length. The kernel allows doing this at any time, but setting the queue length value is not enough. Setting the queue length does not create the transmit queue (going from 0 to non-zero) or destroy it (setting to zero); the queue is only created (or destroyed) when device comes up (or goes down).

This patch handles this be doing the necessary activations.

That seems quite heavy-weight in case of hierarchical qdiscs, it will walk and reset the entire tree and looses all state, including packets as a side effect. Its also a bit inconsistent since it only affects the default pfifos at the root, not the inner default qdiscs if I'm not mistaken.

A less intrusive way would be to simply use dev->tx_queue_len directly in pfifo when no limit is configured. That would require to make a runtime decision for each packet though. Two more alternatives:

- add a special pfifo variant that always uses tx_queue_len. Duplication comes down to pfifo_enqueue (a few lines) and a new struct Qdisc_ops

- if you really only care about the root, reconfigure it explicitly using fifo_set_limit()

Quoted 6 lines
Quoted 2 lines

The problem is when setting limit to 0 to force noop and setting it to non-zero to force pfifo_fast back again.

I'd suggest to simply attach a pfifo qdisc with an explicitly configured limit to activate queueing on virtual devices.

The existing code is broken if user does:

# ip li set dev eth0.200 txq 200
# tc qdisc set dev eth0.200 root sfq
# tc qdisc del dev eth0.200 root
# ip li set dev eth0.200 txq 0

The vlan is now dead because it is impossible to go back to the no queueing (noop) qdisc. Also, when doing performance tests it is useful to be able to turn off transmit queue entirely.

Other alternative would be to reject changes to txq when device is up, but that probably would break existing users. And would be a pain in the ass when trying to configure PPP and tunnel endpoints.

#5 lore raw
Quoted 8 lines · Stephen Hemminger
Quoted 2 lines · On Wed, 18 Mar 2009 19:08:07 +0100 Patrick McHardy <[email protected]> wrote:

I'd suggest to simply attach a pfifo qdisc with an explicitly configured limit to activate queueing on virtual devices.

The existing code is broken if user does:

Quoted 8 lines
# ip li set dev eth0.200 txq 200
# tc qdisc set dev eth0.200 root sfq
# tc qdisc del dev eth0.200 root
# ip li set dev eth0.200 txq 0

The vlan is now dead because it is impossible to go back to the no queueing (noop) qdisc. Also, when doing performance tests it is useful to be able to turn off transmit queue entirely.

What I meant is:

# ip l l dummy0
109: dummy0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN
    link/ether 0a:94:90:3c:e3:45 brd ff:ff:ff:ff:ff:ff
# tc qdisc add dev dummy0 root pfifo limit 1000
# ip l l dummy0
109: dummy0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc pfifo state UNKNOWN
    link/ether 0a:94:90:3c:e3:45 brd ff:ff:ff:ff:ff:ff
# tc qdisc del dev dummy0 root
# ip l l dummy0
109: dummy0: <BROADCAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN
    link/ether 0a:94:90:3c:e3:45 brd ff:ff:ff:ff:ff:ff

It should work your way as well if you set the transmit queue length to zero before removing the root qdisc.

#6 lore raw
In reply to Stephen Hemminger

Some virtual devices like VLAN's or bridges by default have not transmit queue. In order to do queuing disciplines on these devices a queue must be added by setting transmit queue length. The kernel allows doing this at any time, but setting the queue length value is not enough. Setting the queue length does not create the transmit queue (going from 0 to non-zero) or destroy it (setting to zero); the queue is only created (or destroyed) when device comes up (or goes down).

This patch handles this be doing the necessary activations.

---
 include/linux/netdevice.h |    2 ++
 net/core/dev.c            |   19 +++++++++++++++++--
 net/core/net-sysfs.c      |    5 ++---
 net/core/rtnetlink.c      |    2 +-
 4 files changed, 22 insertions(+), 6 deletions(-)
include/linux/netdevice.h +2 -0
@@ -1436,6 +1436,8 @@ extern int		dev_change_net_namespace(str
 extern int		dev_set_mtu(struct net_device *, int);
 extern int		dev_set_mac_address(struct net_device *,
 					    struct sockaddr *);
+extern int		dev_set_tx_queue_len(struct net_device *dev,
+					     unsigned long len);
 extern int		dev_hard_start_xmit(struct sk_buff *skb,
 					    struct net_device *dev,
 					    struct netdev_queue *txq);
net/core/dev.c +25 -2
@@ -983,6 +983,30 @@ int dev_set_alias(struct net_device *dev
 	return len;
 }
 
+/**
+ *
+ *	@dev: device
+ *	@len: limit of frames in transmit queue
+ *
+ *	Set transmit queue length for device.
+ *	If queue length is set to 0, then queueing is disabled.
+ */
+int dev_set_tx_queue_len(struct net_device *dev, unsigned long len)
+{
+	if (len == dev->tx_queue_len)
+		return 0;
+
+	if ((dev->tx_queue_len != 0) && (len != 0))
+		dev->tx_queue_len = len;
+	else {
+		if (dev->flags & IFF_UP)
+			dev_deactivate(dev);
+		dev->tx_queue_len = len;
+		if (dev->flags & IFF_UP)
+			dev_activate(dev);
+	}
+	return 0;
+}
 
 /**
  *	netdev_features_change - device changes features
@@ -3975,8 +3999,7 @@ static int dev_ifsioc(struct net *net, s
 		case SIOCSIFTXQLEN:
 			if (ifr->ifr_qlen < 0)
 				return -EINVAL;
-			dev->tx_queue_len = ifr->ifr_qlen;
-			return 0;
+			return dev_set_tx_queue_len(dev, ifr->ifr_qlen);
 
 		case SIOCSIFNAME:
 			ifr->ifr_newname[IFNAMSIZ-1] = '\0';
net/core/net-sysfs.c +2 -3
@@ -198,10 +198,9 @@ static ssize_t store_flags(struct device
 
 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 
-static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
+static int change_tx_queue_len(struct net_device *netdev, unsigned long new_len)
 {
-	net->tx_queue_len = new_len;
-	return 0;
+	return dev_set_tx_queue_len(netdev, new_len);
 }
 
 static ssize_t store_tx_queue_len(struct device *dev,
net/core/rtnetlink.c +1 -1
@@ -885,7 +885,7 @@ static int do_setlink(struct net_device 
 	}
 
 	if (tb[IFLA_TXQLEN])
-		dev->tx_queue_len = nla_get_u32(tb[IFLA_TXQLEN]);
+		dev_set_tx_queue_len(dev, nla_get_u32(tb[IFLA_TXQLEN]));
 
 	if (tb[IFLA_OPERSTATE])
 		set_operstate(dev, nla_get_u8(tb[IFLA_OPERSTATE]));
#7 lore raw
In reply to Stephen Hemminger
Quoted 33 lines · Stephen Hemminger

Some virtual devices like VLAN's or bridges by default have not transmit queue. In order to do queuing disciplines on these devices a queue must be added by setting transmit queue length. The kernel allows doing this at any time, but setting the queue length value is not enough. Setting the queue length does not create the transmit queue (going from 0 to non-zero) or destroy it (setting to zero); the queue is only created (or destroyed) when device comes up (or goes down).

This patch handles this be doing the necessary activations.

---
 include/linux/netdevice.h |    2 ++
 net/core/dev.c            |   19 +++++++++++++++++--
 net/core/net-sysfs.c      |    5 ++---
 net/core/rtnetlink.c      |    2 +-
 4 files changed, 22 insertions(+), 6 deletions(-)
include/linux/netdevice.h +2 -0
@@ -1436,6 +1436,8 @@ extern int		dev_change_net_namespace(str
 extern int		dev_set_mtu(struct net_device *, int);
 extern int		dev_set_mac_address(struct net_device *,
 					    struct sockaddr *);
+extern int		dev_set_tx_queue_len(struct net_device *dev,
+					     unsigned long len);
 extern int		dev_hard_start_xmit(struct sk_buff *skb,
 					    struct net_device *dev,
 					    struct netdev_queue *txq);

net/core/dev.c int dev_set_alias(struct net_device *dev · line 991

Quoted 9 lines
net/core/dev.c +6 -0
@@ -983,6 +983,22 @@ int dev_set_alias(struct net_device *dev
 	return len;
 }
 
+/**
+ *
+ *	@dev: device
+ *	@len: limit of frames in transmit queue
+ *
+ *	Set ifalias for a device,

copy/pasted from "dev_set_alias - change ifalias of a device" ? ;)

net/core/dev.c int dev_set_alias(struct net_device *dev · line 994

Quoted 3 lines
net/core/dev.c +3 -0
+ */
+int dev_set_tx_queue_len(struct net_device *dev, unsigned long len)
+{

if (dev->tx_queue_len == len)

	return 0;

net/core/dev.c int dev_set_alias(struct net_device *dev · line 996

Quoted 2 lines
net/core/dev.c +2 -0
+	if (dev->flags & IFF_UP)
+		dev_deactivate(dev);

So changing txqueuelen will drop frames ?

Quoted 7 lines
net/core/dev.c +5 -0
+	dev->tx_queue_len = len;
+	if (dev->flags & IFF_UP)
+		dev_activate(dev);
+	return 0;
+}
 
 
#8 lore raw
Quoted 46 lines · On Wed, 18 Mar 2009 18:18:44 +0100 Eric Dumazet <[email protected]> wrote:
Quoted 33 lines · Stephen Hemminger

Some virtual devices like VLAN's or bridges by default have not transmit queue. In order to do queuing disciplines on these devices a queue must be added by setting transmit queue length. The kernel allows doing this at any time, but setting the queue length value is not enough. Setting the queue length does not create the transmit queue (going from 0 to non-zero) or destroy it (setting to zero); the queue is only created (or destroyed) when device comes up (or goes down).

This patch handles this be doing the necessary activations.

---
 include/linux/netdevice.h |    2 ++
 net/core/dev.c            |   19 +++++++++++++++++--
 net/core/net-sysfs.c      |    5 ++---
 net/core/rtnetlink.c      |    2 +-
 4 files changed, 22 insertions(+), 6 deletions(-)
include/linux/netdevice.h +2 -0
@@ -1436,6 +1436,8 @@ extern int		dev_change_net_namespace(str
 extern int		dev_set_mtu(struct net_device *, int);
 extern int		dev_set_mac_address(struct net_device *,
 					    struct sockaddr *);
+extern int		dev_set_tx_queue_len(struct net_device *dev,
+					     unsigned long len);
 extern int		dev_hard_start_xmit(struct sk_buff *skb,
 					    struct net_device *dev,
 					    struct netdev_queue *txq);

net/core/dev.c int dev_set_alias(struct net_device *dev · line 991

Quoted 9 lines
net/core/dev.c +6 -0
@@ -983,6 +983,22 @@ int dev_set_alias(struct net_device *dev
 	return len;
 }
 
+/**
+ *
+ *	@dev: device
+ *	@len: limit of frames in transmit queue
+ *
+ *	Set ifalias for a device,

copy/pasted from "dev_set_alias - change ifalias of a device" ? ;)

Quoted 6 lines

net/core/dev.c int dev_set_alias(struct net_device *dev · line 994

Quoted 3 lines
net/core/dev.c +3 -0
+ */
+int dev_set_tx_queue_len(struct net_device *dev, unsigned long len)
+{

if (dev->tx_queue_len == len)

	return 0;

Also could do:

if ((dev->tx_queue_len != 0) && (len != 0)) {
      dev->tx_queue_len = len;
      return 0;
}