From: Neil Horman <nhorman@redhat.com>
Date: Wed, 8 Sep 2010 17:46:04 -0400
Subject: [net] bonding: enable output slave selection
Message-id: <20100908174604.GE29399@hmsreliant.think-freely.org>
Patchwork-id: 28179
O-Subject: [RHEL5.6 PATCH] bonding: enable output slave selection (bz 516289)
Bugzilla: 516289
RH-Acked-by: David S. Miller <davem@redhat.com>
RH-Acked-by: Jiri Pirko <jpirko@redhat.com>

Hey-
	gospo and I wrote a bonding feature awhile back to do output slave
selection in the bonding driver using the multiqueue tx infrastructure.  Goldman
wanted this ability in RHEL5, and this is a backport of that feature.  Since we
don't have any plans to backport the entire multiqueue tx infrastructure to
RHEL5, this is an approximation of the feature.  Instead of using the traffic
control utility and the the skbedit action to select output queues, we use
iptables and the nfmark field of the skb to select output queues based on slave
ifindex.  The feature is opt-in, enabled by the mark_steering variable in sysfs,
that this patch adds, on a per-bond-interface basis.

Brew status:
https://brewweb.devel.redhat.com/taskinfo?taskID=2748700

Test status:
Verified by myself in dell-pe1950-07.lab.bos.redhat.com

Resolves bz 516289

Signed-off-by: Jarod Wilson <jarod@redhat.com>

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index 586c464..cddd517 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -221,6 +221,8 @@ struct bond_parm_tbl pri_reselect_tbl[] = {
 
 static void bond_send_gratuitous_arp(struct bonding *bond);
 static void bond_deinit(struct net_device *bond_dev);
+static inline int bond_slave_override(struct bonding *bond,
+                                      struct sk_buff *skb);
 
 /*---------------------------- General routines -----------------------------*/
 
@@ -4348,11 +4350,17 @@ static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev
 	int i, slave_no, res = 1;
 	struct iphdr *iph = ip_hdr(skb);
 
+	if (TX_QUEUE_OVERRIDE(bond)) {
+		if (!bond_slave_override(bond, skb))
+			return NETDEV_TX_OK;
+        }
+
 	read_lock(&bond->lock);
 
 	if (!BOND_IS_OK(bond)) {
 		goto out;
 	}
+
 	/*
 	 * Start with the curr_active_slave that joined the bond as the
 	 * default for sending IGMP traffic.  For failover purposes one
@@ -4413,6 +4421,11 @@ static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_d
 	struct bonding *bond = bond_dev->priv;
 	int res = 1;
 
+	if (TX_QUEUE_OVERRIDE(bond)) {
+		if (!bond_slave_override(bond, skb))
+			return NETDEV_TX_OK;
+	}
+
 	read_lock(&bond->lock);
 	read_lock(&bond->curr_slave_lock);
 
@@ -4565,6 +4578,40 @@ static void bond_set_xmit_hash_policy(struct bonding *bond)
 }
 
 /*
+ * Lookup the slave that corresponds to a qid
+ */
+static inline int bond_slave_override(struct bonding *bond,
+				      struct sk_buff *skb)
+{
+	int i, res = 1;
+	struct slave *slave = NULL;
+	struct slave *check_slave;
+
+	read_lock(&bond->lock);
+
+	if (!BOND_IS_OK(bond) || !skb->nfmark)
+		goto out;
+
+	/* Find out if any slaves have the same mapping as this skb. */
+	bond_for_each_slave(bond, check_slave, i) {
+		if (check_slave->dev->ifindex == skb->nfmark) {
+			slave = check_slave;
+			break;
+		}
+	}
+
+	/* If the slave isn't UP, use default transmit policy. */
+	if (slave && IS_UP(slave->dev) &&
+	    (slave->link == BOND_LINK_UP)) {
+		res = bond_dev_queue_xmit(bond, skb, slave->dev);
+	}
+
+out:
+	read_unlock(&bond->lock);
+	return res;
+}
+
+/*
  * set bond mode specific net device operations
  */
 void bond_set_mode_ops(struct bonding *bond, int mode)
diff --git a/drivers/net/bonding/bond_sysfs.c b/drivers/net/bonding/bond_sysfs.c
index d6bf50f..977ae5a 100644
--- a/drivers/net/bonding/bond_sysfs.c
+++ b/drivers/net/bonding/bond_sysfs.c
@@ -1437,6 +1437,34 @@ static ssize_t bonding_show_ad_partner_mac(struct class_device *cd, char *buf)
 }
 static CLASS_DEVICE_ATTR(ad_partner_mac, S_IRUGO, bonding_show_ad_partner_mac, NULL);
 
+static ssize_t bonding_store_mark_steering(struct class_device *cd,
+					   const char *buf, size_t count)
+{
+	int val;
+	int rc;
+	struct bonding *bond = to_bond(cd);
+
+	rc = sscanf(buf, "%d", &val);
+	if (rc != 1)
+		return -EINVAL;
+
+	bond->mark_steering = (val) ? true:false;
+	
+	return count;
+}
+
+static ssize_t bonding_show_mark_steering(struct class_device *cd, char *buf)
+{
+	struct bonding *bond = to_bond(cd);
+	int count = 0;
+
+	count = sprintf(buf, "%d\n", (bond->mark_steering == true) ? 1:0);
+
+	return count;
+}
+
+static CLASS_DEVICE_ATTR(mark_steering, S_IWUSR | S_IRUGO, bonding_show_mark_steering,
+			 bonding_store_mark_steering);
 
 
 static struct attribute *per_bond_attrs[] = {
@@ -1463,6 +1491,7 @@ static struct attribute *per_bond_attrs[] = {
 	&class_device_attr_ad_actor_key.attr,
 	&class_device_attr_ad_partner_key.attr,
 	&class_device_attr_ad_partner_mac.attr,
+	&class_device_attr_mark_steering.attr,
 	NULL,
 };
 
diff --git a/drivers/net/bonding/bonding.h b/drivers/net/bonding/bonding.h
index 1c9c6db..cddc6d4 100644
--- a/drivers/net/bonding/bonding.h
+++ b/drivers/net/bonding/bonding.h
@@ -68,6 +68,8 @@ extern int debug;
 		 ((mode) == BOND_MODE_TLB)          ||	\
 		 ((mode) == BOND_MODE_ALB))
 
+#define TX_QUEUE_OVERRIDE(bond) (bond->mark_steering == true) 
+
 /*
  * Less bad way to call ioctl from within the kernel; this needs to be
  * done some other way to get the call out of interrupt context.
@@ -202,6 +204,7 @@ struct bonding {
 	struct   slave *current_arp_slave;
 	struct   slave *primary_slave;
 	bool     force_primary;
+	bool     mark_steering;
 	s32      slave_cnt; /* never change this value outside the attach/detach wrappers */
 	rwlock_t lock;
 	rwlock_t curr_slave_lock;