Skip to contentSkip to conversation

net/ipv6: Block IPv6 addrconf on team ports

PATCH In review Chas Williams 3 participants 7 messages Oct 26, 2018

Subsystems (5)
Networking Network drivers IPv4/IPv6 Bonding driver +1 more subsystems
lore
Share & tools
lore

Chas Williams

author
#0 lore raw

PATCH 1/1 net/ipv6: Block IPv6 addrconf on team ports

netif_is_lag_port should be used to identify link aggregation ports. For this to work, we need to reorganize the bonding and team drivers so that the necessary flags are set before dev_open is called.

commit 31e77c93e432 ("sched/fair: Update blocked load when newly idle") made this decision originally based on the IFF_SLAVE flag which isn't used by the team driver. Note, we do need to retain the IFF_SLAVE check for the eql driver.

Signed-off-by Chas Williams <[email protected]>
---
 drivers/net/bonding/bond_main.c | 4 ++--
 drivers/net/team/team.c         | 7 +++++--
 net/ipv6/addrconf.c             | 2 +-
 3 files changed, 8 insertions(+), 5 deletions(-)
drivers/net/bonding/bond_main.c +2 -2
index ffa37adb7681..5cdad164332b 100644
@@ -1536,6 +1536,7 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 
 	/* set slave flag before open to prevent IPv6 addrconf */
 	slave_dev->flags |= IFF_SLAVE;
+	slave_dev->priv_flags |= IFF_BONDING;
 
 	/* open the slave since the application closed it */
 	res = dev_open(slave_dev);

@@ -1544,7 +1545,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 		goto err_restore_mac;
 	}
 
-	slave_dev->priv_flags |= IFF_BONDING;
 	/* initialize slave stats */
 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
 

@@ -1804,10 +1804,10 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 	slave_disable_netpoll(new_slave);
 
 err_close:
-	slave_dev->priv_flags &= ~IFF_BONDING;
 	dev_close(slave_dev);
 
 err_restore_mac:
+	slave_dev->priv_flags &= ~IFF_BONDING;
 	slave_dev->flags &= ~IFF_SLAVE;
 	if (!bond->params.fail_over_mac ||
 	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
drivers/net/team/team.c +5 -2
index db633ae9f784..8fc7d57e9f6d 100644
@@ -1128,14 +1128,12 @@ static int team_upper_dev_link(struct team *team, struct team_port *port,
 					   &lag_upper_info, extack);
 	if (err)
 		return err;
-	port->dev->priv_flags |= IFF_TEAM_PORT;
 	return 0;
 }
 
 static void team_upper_dev_unlink(struct team *team, struct team_port *port)
 {
 	netdev_upper_dev_unlink(port->dev, team->dev);
-	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 }
 
 static void __team_port_change_port_added(struct team_port *port, bool linkup);

@@ -1214,6 +1212,9 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
 		goto err_port_enter;
 	}
 
+	/* set slave flag before open to prevent IPv6 addrconf */
+	port->dev->priv_flags |= IFF_TEAM_PORT;
+
 	err = dev_open(port_dev);
 	if (err) {
 		netdev_dbg(dev, "Device %s opening failed\n",

@@ -1292,6 +1293,7 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
 	dev_close(port_dev);
 
 err_dev_open:
+	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 	team_port_leave(team, port);
 	team_port_set_orig_dev_addr(port);
 

@@ -1328,6 +1330,7 @@ static int team_port_del(struct team *team, struct net_device *port_dev)
 	dev_uc_unsync(port_dev, dev);
 	dev_mc_unsync(port_dev, dev);
 	dev_close(port_dev);
+	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 	team_port_leave(team, port);
 
 	__team_option_inst_mark_removed_port(team, port);
net/ipv6/addrconf.c +1 -1
index 45b84dd5c4eb..121f863022ed 100644
@@ -3482,7 +3482,7 @@ static int addrconf_notify(struct notifier_block *this, unsigned long event,
 
 	case NETDEV_UP:
 	case NETDEV_CHANGE:
-		if (dev->flags & IFF_SLAVE)
+		if (netif_is_lag_port(dev) || dev->flags & IFF_SLAVE)
 			break;
 
 		if (idev && idev->cnf.disable_ipv6)
Signature · 1 line
2.14.4
maintainer
#1 lore raw
Quoted 9 lines · Thu, Oct 25, 2018 at 11:02:27PM CEST, [email protected] wrote:

netif_is_lag_port should be used to identify link aggregation ports. For this to work, we need to reorganize the bonding and team drivers so that the necessary flags are set before dev_open is called.

commit 31e77c93e432 ("sched/fair: Update blocked load when newly idle") made this decision originally based on the IFF_SLAVE flag which isn't used by the team driver. Note, we do need to retain the IFF_SLAVE check for the eql driver.

Quoted 5 lines
Signed-off-by Chas Williams <[email protected]>

---

drivers/net/bonding/bond_main.c | 4 ++--
drivers/net/team/team.c         | 7 +++++--
net/ipv6/addrconf.c             | 2 +-

Subject talks about "team" yet you modify bond and team. Confusing..

author
#2 lore raw
Quoted 10 lines · Jiri Pirko
Quoted 9 lines · Thu, Oct 25, 2018 at 11:02:27PM CEST, [email protected] wrote:

netif_is_lag_port should be used to identify link aggregation ports. For this to work, we need to reorganize the bonding and team drivers so that the necessary flags are set before dev_open is called.

commit 31e77c93e432 ("sched/fair: Update blocked load when newly idle") made this decision originally based on the IFF_SLAVE flag which isn't used by the team driver. Note, we do need to retain the IFF_SLAVE check for the eql driver.

Quoted 7 lines
Quoted 5 lines
Signed-off-by Chas Williams <[email protected]>

---

drivers/net/bonding/bond_main.c | 4 ++--
drivers/net/team/team.c         | 7 +++++--
net/ipv6/addrconf.c             | 2 +-

Subject talks about "team" yet you modify bond and team. Confusing..

The subject discusses what I want to do. The body of the message covers how I had to do it. The behavior of bonding with respect to addrconf isn't changed but netif_is_lag_port is picky about the flags it wants to see from bonding. So some bonding changes are necessary.

#3 lore raw
In reply to Chas Williams
Quoted 8 lines · Chas Williams

netif_is_lag_port should be used to identify link aggregation ports. For this to work, we need to reorganize the bonding and team drivers so that the necessary flags are set before dev_open is called.

commit 31e77c93e432 ("sched/fair: Update blocked load when newly idle") made this decision originally based on the IFF_SLAVE flag which isn't used by the team driver. Note, we do need to retain the IFF_SLAVE check for the eql driver.

Is 31e77c93e432 the correct commit reference? I don't see
anything in there about IFF_SLAVE or bonding; it's a patch to the process scheduler.

And, as Jiri said, the subject doesn't mention bonding.

Quoted 89 lines
Signed-off-by Chas Williams <[email protected]>

---

drivers/net/bonding/bond_main.c | 4 ++--
drivers/net/team/team.c         | 7 +++++--
net/ipv6/addrconf.c             | 2 +-
3 files changed, 8 insertions(+), 5 deletions(-)
drivers/net/bonding/bond_main.c +1 -1
index ffa37adb7681..5cdad164332b 100644
@@ -1536,6 +1536,7 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 
 	/* set slave flag before open to prevent IPv6 addrconf */
 	slave_dev->flags |= IFF_SLAVE;
+	slave_dev->priv_flags |= IFF_BONDING;
 
 	/* open the slave since the application closed it */
 	res = dev_open(slave_dev);

@@ -1544,7 +1545,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 		goto err_restore_mac;
 	}
 
-	slave_dev->priv_flags |= IFF_BONDING;
 	/* initialize slave stats */
 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
 

@@ -1804,10 +1804,10 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 	slave_disable_netpoll(new_slave);

err_close:

drivers/net/bonding/bond_main.c +0 -1
-	slave_dev->priv_flags &= ~IFF_BONDING;

	dev_close(slave_dev);

err_restore_mac:

drivers/net/bonding/bond_main.c +1 -0
+	slave_dev->priv_flags &= ~IFF_BONDING;

	slave_dev->flags &= ~IFF_SLAVE;
	if (!bond->params.fail_over_mac ||
	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
drivers/net/team/team.c +3 -2
index db633ae9f784..8fc7d57e9f6d 100644
@@ -1128,14 +1128,12 @@ static int team_upper_dev_link(struct team *team, struct team_port *port,
 					   &lag_upper_info, extack);
 	if (err)
 		return err;
-	port->dev->priv_flags |= IFF_TEAM_PORT;
 	return 0;
 }
 
 static void team_upper_dev_unlink(struct team *team, struct team_port *port)
 {
 	netdev_upper_dev_unlink(port->dev, team->dev);
-	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 }
 
 static void __team_port_change_port_added(struct team_port *port, bool linkup);

@@ -1214,6 +1212,9 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
 		goto err_port_enter;
 	}
 
+	/* set slave flag before open to prevent IPv6 addrconf */
+	port->dev->priv_flags |= IFF_TEAM_PORT;
+
 	err = dev_open(port_dev);
 	if (err) {
 		netdev_dbg(dev, "Device %s opening failed\n",

@@ -1292,6 +1293,7 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
 	dev_close(port_dev);

err_dev_open:

drivers/net/team/team.c +1 -0
+	port->dev->priv_flags &= ~IFF_TEAM_PORT;

	team_port_leave(team, port);
	team_port_set_orig_dev_addr(port);
drivers/net/team/team.c +1 -0
@@ -1328,6 +1330,7 @@ static int team_port_del(struct team *team, struct net_device *port_dev)
 	dev_uc_unsync(port_dev, dev);
 	dev_mc_unsync(port_dev, dev);
 	dev_close(port_dev);
+	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 	team_port_leave(team, port);
 
 	__team_option_inst_mark_removed_port(team, port);
net/ipv6/addrconf.c +0 -0
index 45b84dd5c4eb..121f863022ed 100644

net/ipv6/addrconf.c static int addrconf_notify(struct notifier_block *this, unsigned long event, · line 3485

Quoted 5 lines
net/ipv6/addrconf.c +1 -1
@@ -3482,7 +3482,7 @@ static int addrconf_notify(struct notifier_block *this, unsigned long event,
 
 	case NETDEV_UP:
 	case NETDEV_CHANGE:
-		if (dev->flags & IFF_SLAVE)
+		if (netif_is_lag_port(dev) || dev->flags & IFF_SLAVE)

Note that netvsc_vf_join() also uses IFF_SLAVE in order skip
IPv6 addrconf for netvsc devices; I don't believe its usage will pass netif_is_lag_port(). It looks like the above will work, but your commit message mentions eql as the reason for retaining the IFF_SLAVE test, and eql isn't the only user of IFF_SLAVE in this manner.

	-J
Quoted 6 lines
net/ipv6/addrconf.c +0 -0
 			break;
 
 		if (idev && idev->cnf.disable_ipv6)
Signature · 1 line
2.14.4

---
-Jay Vosburgh, [email protected]

author
#4 lore raw
Quoted 6 lines · Jay Vosburgh
Quoted 4 lines · Chas Williams

netif_is_lag_port should be used to identify link aggregation ports. For this to work, we need to reorganize the bonding and team drivers so that the necessary flags are set before dev_open is called.

Quoted 8 lines
Quoted 4 lines

commit 31e77c93e432 ("sched/fair: Update blocked load when newly idle") made this decision originally based on the IFF_SLAVE flag which isn't used by the team driver. Note, we do need to retain the IFF_SLAVE check for the eql driver.

Is 31e77c93e432 the correct commit reference? I don't see
anything in there about IFF_SLAVE or bonding; it's a patch to the process scheduler.

No, that's wrong. It should be c2edacf80e155.

Quoted 1 line

And, as Jiri said, the subject doesn't mention bonding.

The behavior of bonding wasn't changed. The intent of the patch is to add team slaves to the interfaces that don't get automatic IPv6 addresses. The body discusses why bonding had to change as well.

I was under the impression that the subject needs to kept short. If there a better way to phrase what I want to do?

Quoted 114 lines
Quoted 90 lines
Signed-off-by Chas Williams <[email protected]>

---

drivers/net/bonding/bond_main.c | 4 ++--
drivers/net/team/team.c         | 7 +++++--
net/ipv6/addrconf.c             | 2 +-
3 files changed, 8 insertions(+), 5 deletions(-)
drivers/net/bonding/bond_main.c +1 -1
index ffa37adb7681..5cdad164332b 100644
@@ -1536,6 +1536,7 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 
 	/* set slave flag before open to prevent IPv6 addrconf */
 	slave_dev->flags |= IFF_SLAVE;
+	slave_dev->priv_flags |= IFF_BONDING;
 
 	/* open the slave since the application closed it */
 	res = dev_open(slave_dev);

@@ -1544,7 +1545,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 		goto err_restore_mac;
 	}
 
-	slave_dev->priv_flags |= IFF_BONDING;
 	/* initialize slave stats */
 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
 

@@ -1804,10 +1804,10 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 	slave_disable_netpoll(new_slave);

err_close:

drivers/net/bonding/bond_main.c +0 -1
-	slave_dev->priv_flags &= ~IFF_BONDING;

	dev_close(slave_dev);

err_restore_mac:

drivers/net/bonding/bond_main.c +1 -0
+	slave_dev->priv_flags &= ~IFF_BONDING;

	slave_dev->flags &= ~IFF_SLAVE;
	if (!bond->params.fail_over_mac ||
	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
drivers/net/team/team.c +3 -2
index db633ae9f784..8fc7d57e9f6d 100644
@@ -1128,14 +1128,12 @@ static int team_upper_dev_link(struct team *team, struct team_port *port,
 					   &lag_upper_info, extack);
 	if (err)
 		return err;
-	port->dev->priv_flags |= IFF_TEAM_PORT;
 	return 0;
 }
 
 static void team_upper_dev_unlink(struct team *team, struct team_port *port)
 {
 	netdev_upper_dev_unlink(port->dev, team->dev);
-	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 }
 
 static void __team_port_change_port_added(struct team_port *port, bool linkup);

@@ -1214,6 +1212,9 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
 		goto err_port_enter;
 	}
 
+	/* set slave flag before open to prevent IPv6 addrconf */
+	port->dev->priv_flags |= IFF_TEAM_PORT;
+
 	err = dev_open(port_dev);
 	if (err) {
 		netdev_dbg(dev, "Device %s opening failed\n",

@@ -1292,6 +1293,7 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
 	dev_close(port_dev);

err_dev_open:

drivers/net/team/team.c +1 -0
+	port->dev->priv_flags &= ~IFF_TEAM_PORT;

	team_port_leave(team, port);
	team_port_set_orig_dev_addr(port);
drivers/net/team/team.c +1 -0
@@ -1328,6 +1330,7 @@ static int team_port_del(struct team *team, struct net_device *port_dev)
 	dev_uc_unsync(port_dev, dev);
 	dev_mc_unsync(port_dev, dev);
 	dev_close(port_dev);
+	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 	team_port_leave(team, port);
 
 	__team_option_inst_mark_removed_port(team, port);
net/ipv6/addrconf.c +0 -0
index 45b84dd5c4eb..121f863022ed 100644

net/ipv6/addrconf.c static int addrconf_notify(struct notifier_block *this, unsigned long event, · line 3485

Quoted 5 lines
net/ipv6/addrconf.c +1 -1
@@ -3482,7 +3482,7 @@ static int addrconf_notify(struct notifier_block *this, unsigned long event,
 
 	case NETDEV_UP:
 	case NETDEV_CHANGE:
-		if (dev->flags & IFF_SLAVE)
+		if (netif_is_lag_port(dev) || dev->flags & IFF_SLAVE)

Note that netvsc_vf_join() also uses IFF_SLAVE in order skip
IPv6 addrconf for netvsc devices; I don't believe its usage will pass netif_is_lag_port(). It looks like the above will work, but your commit message mentions eql as the reason for retaining the IFF_SLAVE test, and eql isn't the only user of IFF_SLAVE in this manner.

	-J
Quoted 6 lines
net/ipv6/addrconf.c +0 -0
 			break;
 
 		if (idev && idev->cnf.disable_ipv6)
Signature · 1 line
2.14.4

---
-Jay Vosburgh, [email protected]

#5 lore raw
Quoted 16 lines · Chas Williams
Quoted 6 lines · Jay Vosburgh
Quoted 4 lines · Chas Williams

netif_is_lag_port should be used to identify link aggregation ports. For this to work, we need to reorganize the bonding and team drivers so that the necessary flags are set before dev_open is called.

Quoted 8 lines
Quoted 4 lines

commit 31e77c93e432 ("sched/fair: Update blocked load when newly idle") made this decision originally based on the IFF_SLAVE flag which isn't used by the team driver. Note, we do need to retain the IFF_SLAVE check for the eql driver.

Is 31e77c93e432 the correct commit reference? I don't see
anything in there about IFF_SLAVE or bonding; it's a patch to the process scheduler.

Quoted 8 lines

No, that's wrong. It should be c2edacf80e155.

Quoted 1 line

And, as Jiri said, the subject doesn't mention bonding.

The behavior of bonding wasn't changed. The intent of the patch is to add team slaves to the interfaces that don't get automatic IPv6 addresses. The body discusses why bonding had to change as well.

Sure, but the bonding code has changed, and the current
presentation makes it harder for reviewers to follow (or perhaps even notice).

Quoted 2 lines

I was under the impression that the subject needs to kept short. If there a better way to phrase what I want to do?

I'd suggest splitting this into three patches: A first patch
that adds the new IPv6 functionality, then one patch each for team and bonding to take advantage of that new functionality. Each of the three would then be very straightforward, change just one thing, and should be clearer all around.

	-J
Quoted 101 lines
Quoted 101 lines
Quoted 89 lines
Signed-off-by Chas Williams <[email protected]>

---

drivers/net/bonding/bond_main.c | 4 ++--
drivers/net/team/team.c         | 7 +++++--
net/ipv6/addrconf.c             | 2 +-
3 files changed, 8 insertions(+), 5 deletions(-)
drivers/net/bonding/bond_main.c +1 -1
index ffa37adb7681..5cdad164332b 100644
@@ -1536,6 +1536,7 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 
 	/* set slave flag before open to prevent IPv6 addrconf */
 	slave_dev->flags |= IFF_SLAVE;
+	slave_dev->priv_flags |= IFF_BONDING;
 
 	/* open the slave since the application closed it */
 	res = dev_open(slave_dev);

@@ -1544,7 +1545,6 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 		goto err_restore_mac;
 	}
 
-	slave_dev->priv_flags |= IFF_BONDING;
 	/* initialize slave stats */
 	dev_get_stats(new_slave->dev, &new_slave->slave_stats);
 

@@ -1804,10 +1804,10 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 	slave_disable_netpoll(new_slave);

err_close:

drivers/net/bonding/bond_main.c +0 -1
-	slave_dev->priv_flags &= ~IFF_BONDING;

	dev_close(slave_dev);

err_restore_mac:

drivers/net/bonding/bond_main.c +1 -0
+	slave_dev->priv_flags &= ~IFF_BONDING;

	slave_dev->flags &= ~IFF_SLAVE;
	if (!bond->params.fail_over_mac ||
	    BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
drivers/net/team/team.c +3 -2
index db633ae9f784..8fc7d57e9f6d 100644
@@ -1128,14 +1128,12 @@ static int team_upper_dev_link(struct team *team, struct team_port *port,
 					   &lag_upper_info, extack);
 	if (err)
 		return err;
-	port->dev->priv_flags |= IFF_TEAM_PORT;
 	return 0;
 }
 
 static void team_upper_dev_unlink(struct team *team, struct team_port *port)
 {
 	netdev_upper_dev_unlink(port->dev, team->dev);
-	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 }
 
 static void __team_port_change_port_added(struct team_port *port, bool linkup);

@@ -1214,6 +1212,9 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
 		goto err_port_enter;
 	}
 
+	/* set slave flag before open to prevent IPv6 addrconf */
+	port->dev->priv_flags |= IFF_TEAM_PORT;
+
 	err = dev_open(port_dev);
 	if (err) {
 		netdev_dbg(dev, "Device %s opening failed\n",

@@ -1292,6 +1293,7 @@ static int team_port_add(struct team *team, struct net_device *port_dev,
 	dev_close(port_dev);

err_dev_open:

drivers/net/team/team.c +1 -0
+	port->dev->priv_flags &= ~IFF_TEAM_PORT;

	team_port_leave(team, port);
	team_port_set_orig_dev_addr(port);
drivers/net/team/team.c +1 -0
@@ -1328,6 +1330,7 @@ static int team_port_del(struct team *team, struct net_device *port_dev)
 	dev_uc_unsync(port_dev, dev);
 	dev_mc_unsync(port_dev, dev);
 	dev_close(port_dev);
+	port->dev->priv_flags &= ~IFF_TEAM_PORT;
 	team_port_leave(team, port);
 
 	__team_option_inst_mark_removed_port(team, port);
net/ipv6/addrconf.c +0 -0
index 45b84dd5c4eb..121f863022ed 100644

net/ipv6/addrconf.c static int addrconf_notify(struct notifier_block *this, unsigned long event, · line 3485

Quoted 5 lines
net/ipv6/addrconf.c +1 -1
@@ -3482,7 +3482,7 @@ static int addrconf_notify(struct notifier_block *this, unsigned long event,
 
 	case NETDEV_UP:
 	case NETDEV_CHANGE:
-		if (dev->flags & IFF_SLAVE)
+		if (netif_is_lag_port(dev) || dev->flags & IFF_SLAVE)

Note that netvsc_vf_join() also uses IFF_SLAVE in order skip
IPv6 addrconf for netvsc devices; I don't believe its usage will pass netif_is_lag_port(). It looks like the above will work, but your commit message mentions eql as the reason for retaining the IFF_SLAVE test, and eql isn't the only user of IFF_SLAVE in this manner.

Quoted 7 lines
Quoted 7 lines
	-J
Quoted 5 lines
net/ipv6/addrconf.c +0 -0
 			break;
 
 		if (idev && idev->cnf.disable_ipv6)
Signature · 1 line
2.14.4

---
-Jay Vosburgh, [email protected]

maintainer
#6 lore raw
Quoted 31 lines · Fri, Oct 26, 2018 at 12:40:31AM CEST, [email protected] wrote:
Quoted 16 lines · Chas Williams
Quoted 6 lines · Jay Vosburgh
Quoted 4 lines · Chas Williams

netif_is_lag_port should be used to identify link aggregation ports. For this to work, we need to reorganize the bonding and team drivers so that the necessary flags are set before dev_open is called.

Quoted 8 lines
Quoted 4 lines

commit 31e77c93e432 ("sched/fair: Update blocked load when newly idle") made this decision originally based on the IFF_SLAVE flag which isn't used by the team driver. Note, we do need to retain the IFF_SLAVE check for the eql driver.

Is 31e77c93e432 the correct commit reference? I don't see
anything in there about IFF_SLAVE or bonding; it's a patch to the process scheduler.

Quoted 8 lines

No, that's wrong. It should be c2edacf80e155.

Quoted 1 line

And, as Jiri said, the subject doesn't mention bonding.

The behavior of bonding wasn't changed. The intent of the patch is to add team slaves to the interfaces that don't get automatic IPv6 addresses. The body discusses why bonding had to change as well.

Sure, but the bonding code has changed, and the current
presentation makes it harder for reviewers to follow (or perhaps even notice).

Quoted 8 lines
Quoted 2 lines

I was under the impression that the subject needs to kept short. If there a better way to phrase what I want to do?

I'd suggest splitting this into three patches: A first patch
that adds the new IPv6 functionality, then one patch each for team and bonding to take advantage of that new functionality. Each of the three would then be very straightforward, change just one thing, and should be clearer all around.

+1