Merge branch 'bonding-initial-support-for-hardware-crypto-offload'

Jarod Wilson says:

====================
bonding: initial support for hardware crypto offload

This is an initial functional implementation for doing pass-through of
hardware encryption from bonding device to capable slaves, in active-backup
bond setups. This was developed and tested using ixgbe-driven Intel x520
interfaces with libreswan and a transport mode connection, primarily using
netperf, with assorted connection failures forced during transmission. The
failover works quite well in my testing, and overall performance is right
on par with offload when running on a bare interface, no bond involved.

Caveats: this is ONLY enabled for active-backup, because I'm not sure
how one would manage multiple offload handles for different devices all
running at the same time in the same xfrm, and it relies on some minor
changes to both the xfrm code and slave device driver code to get things
to behave, and I don't have immediate access to any other hardware that
could function similarly, but the NIC driver changes are minimal and
straight-forward enough that I've included what I think ought to be
enough for mlx5 devices too.

v2: reordered patches, switched (back) to using CONFIG_XFRM_OFFLOAD
to wrap the code additions and wrapped overlooked additions.
v3: rebase w/net-next open, add proper cc list to cover letter
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/bonding/bond_main.c

@@ -79,6 +79,7 @@
 #include <net/pkt_sched.h>
 #include <linux/rculist.h>
 #include <net/flow_dissector.h>
+#include <net/xfrm.h>
 #include <net/bonding.h>
 #include <net/bond_3ad.h>
 #include <net/bond_alb.h>
@@ -278,8 +279,6 @@ const char *bond_mode_name(int mode)
 	return names[mode];
 }
 
-/*---------------------------------- VLAN -----------------------------------*/
-
 /**
  * bond_dev_queue_xmit - Prepare skb for xmit.
  *
@@ -302,6 +301,8 @@ netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 	return dev_queue_xmit(skb);
 }
 
+/*---------------------------------- VLAN -----------------------------------*/
+
 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
  * We don't protect the slave list iteration with a lock because:
  * a. This operation is performed in IOCTL context,
@@ -372,6 +373,84 @@ static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 	return 0;
 }
 
+/*---------------------------------- XFRM -----------------------------------*/
+
+#ifdef CONFIG_XFRM_OFFLOAD
+/**
+ * bond_ipsec_add_sa - program device with a security association
+ * @xs: pointer to transformer state struct
+ **/
+static int bond_ipsec_add_sa(struct xfrm_state *xs)
+{
+	struct net_device *bond_dev = xs->xso.dev;
+	struct bonding *bond = netdev_priv(bond_dev);
+	struct slave *slave = rtnl_dereference(bond->curr_active_slave);
+
+	xs->xso.slave_dev = slave->dev;
+	bond->xs = xs;
+
+	if (!(slave->dev->xfrmdev_ops
+	      && slave->dev->xfrmdev_ops->xdo_dev_state_add)) {
+		slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
+		return -EINVAL;
+	}
+
+	return slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
+}
+
+/**
+ * bond_ipsec_del_sa - clear out this specific SA
+ * @xs: pointer to transformer state struct
+ **/
+static void bond_ipsec_del_sa(struct xfrm_state *xs)
+{
+	struct net_device *bond_dev = xs->xso.dev;
+	struct bonding *bond = netdev_priv(bond_dev);
+	struct slave *slave = rtnl_dereference(bond->curr_active_slave);
+
+	if (!slave)
+		return;
+
+	xs->xso.slave_dev = slave->dev;
+
+	if (!(slave->dev->xfrmdev_ops
+	      && slave->dev->xfrmdev_ops->xdo_dev_state_delete)) {
+		slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
+		return;
+	}
+
+	slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
+}
+
+/**
+ * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
+ * @skb: current data packet
+ * @xs: pointer to transformer state struct
+ **/
+static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
+{
+	struct net_device *bond_dev = xs->xso.dev;
+	struct bonding *bond = netdev_priv(bond_dev);
+	struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
+	struct net_device *slave_dev = curr_active->dev;
+
+	if (!(slave_dev->xfrmdev_ops
+	      && slave_dev->xfrmdev_ops->xdo_dev_offload_ok)) {
+		slave_warn(bond_dev, slave_dev, "%s: no slave xdo_dev_offload_ok\n", __func__);
+		return false;
+	}
+
+	xs->xso.slave_dev = slave_dev;
+	return slave_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
+}
+
+static const struct xfrmdev_ops bond_xfrmdev_ops = {
+	.xdo_dev_state_add = bond_ipsec_add_sa,
+	.xdo_dev_state_delete = bond_ipsec_del_sa,
+	.xdo_dev_offload_ok = bond_ipsec_offload_ok,
+};
+#endif /* CONFIG_XFRM_OFFLOAD */
+
 /*------------------------------- Link status -------------------------------*/
 
 /* Set the carrier state for the master according to the state of its
@@ -879,6 +958,11 @@ void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 		return;
 
 	if (new_active) {
+#ifdef CONFIG_XFRM_OFFLOAD
+		if ((BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) && bond->xs)
+			bond_ipsec_del_sa(bond->xs);
+#endif /* CONFIG_XFRM_OFFLOAD */
+
 		new_active->last_link_up = jiffies;
 
 		if (new_active->link == BOND_LINK_BACK) {
@@ -941,6 +1025,13 @@ void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 					bond_should_notify_peers(bond);
 			}
 
+#ifdef CONFIG_XFRM_OFFLOAD
+			if (old_active && bond->xs) {
+				xfrm_dev_state_flush(dev_net(bond->dev), bond->dev, true);
+				bond_ipsec_add_sa(bond->xs);
+			}
+#endif /* CONFIG_XFRM_OFFLOAD */
+
 			call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 			if (should_notify_peers) {
 				bond->send_peer_notif--;
@@ -1127,15 +1218,24 @@ static netdev_features_t bond_fix_features(struct net_device *dev,
 #define BOND_ENC_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
 				 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
 
+#ifdef CONFIG_XFRM_OFFLOAD
+#define BOND_XFRM_FEATURES	(NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM | \
+				 NETIF_F_GSO_ESP)
+#endif /* CONFIG_XFRM_OFFLOAD */
+
 #define BOND_MPLS_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
 				 NETIF_F_ALL_TSO)
 
+
 static void bond_compute_features(struct bonding *bond)
 {
 	unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
 					IFF_XMIT_DST_RELEASE_PERM;
 	netdev_features_t vlan_features = BOND_VLAN_FEATURES;
 	netdev_features_t enc_features  = BOND_ENC_FEATURES;
+#ifdef CONFIG_XFRM_OFFLOAD
+	netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
+#endif /* CONFIG_XFRM_OFFLOAD */
 	netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
 	struct net_device *bond_dev = bond->dev;
 	struct list_head *iter;
@@ -1157,6 +1257,12 @@ static void bond_compute_features(struct bonding *bond)
 							 slave->dev->hw_enc_features,
 							 BOND_ENC_FEATURES);
 
+#ifdef CONFIG_XFRM_OFFLOAD
+		xfrm_features = netdev_increment_features(xfrm_features,
+							  slave->dev->hw_enc_features,
+							  BOND_XFRM_FEATURES);
+#endif /* CONFIG_XFRM_OFFLOAD */
+
 		mpls_features = netdev_increment_features(mpls_features,
 							  slave->dev->mpls_features,
 							  BOND_MPLS_FEATURES);
@@ -1176,6 +1282,9 @@ static void bond_compute_features(struct bonding *bond)
 				    NETIF_F_HW_VLAN_CTAG_TX |
 				    NETIF_F_HW_VLAN_STAG_TX |
 				    NETIF_F_GSO_UDP_L4;
+#ifdef CONFIG_XFRM_OFFLOAD
+	bond_dev->hw_enc_features |= xfrm_features;
+#endif /* CONFIG_XFRM_OFFLOAD */
 	bond_dev->mpls_features = mpls_features;
 	bond_dev->gso_max_segs = gso_max_segs;
 	netif_set_gso_max_size(bond_dev, gso_max_size);
@@ -1464,6 +1573,9 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
 		slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
 	}
 
+	if (slave_dev->features & NETIF_F_HW_ESP)
+		slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
+
 	/* Old ifenslave binaries are no longer supported.  These can
 	 * be identified with moderate accuracy by the state of the slave:
 	 * the current ifenslave will set the interface down prior to
@@ -4540,6 +4652,13 @@ void bond_setup(struct net_device *bond_dev)
 	bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
 	bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
 
+#ifdef CONFIG_XFRM_OFFLOAD
+	/* set up xfrm device ops (only supported in active-backup right now) */
+	if ((BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP))
+		bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
+	bond->xs = NULL;
+#endif /* CONFIG_XFRM_OFFLOAD */
+
 	/* don't acquire bond device's netif_tx_lock when transmitting */
 	bond_dev->features |= NETIF_F_LLTX;
 
@@ -4558,6 +4677,10 @@ void bond_setup(struct net_device *bond_dev)
 				NETIF_F_HW_VLAN_CTAG_FILTER;
 
 	bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
+#ifdef CONFIG_XFRM_OFFLOAD
+	if ((BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP))
+		bond_dev->hw_features |= BOND_XFRM_FEATURES;
+#endif /* CONFIG_XFRM_OFFLOAD */
 	bond_dev->features |= bond_dev->hw_features;
 	bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
 }

drivers/net/ethernet/intel/ixgbe/ixgbe_ipsec.c

@@ -432,6 +432,9 @@ static int ixgbe_ipsec_parse_proto_keys(struct xfrm_state *xs,
 	char *alg_name = NULL;
 	int key_len;
 
+	if (xs->xso.slave_dev)
+		dev = xs->xso.slave_dev;
+
 	if (!xs->aead) {
 		netdev_err(dev, "Unsupported IPsec algorithm\n");
 		return -EINVAL;
@@ -478,8 +481,8 @@ static int ixgbe_ipsec_parse_proto_keys(struct xfrm_state *xs,
 static int ixgbe_ipsec_check_mgmt_ip(struct xfrm_state *xs)
 {
 	struct net_device *dev = xs->xso.dev;
-	struct ixgbe_adapter *adapter = netdev_priv(dev);
-	struct ixgbe_hw *hw = &adapter->hw;
+	struct ixgbe_adapter *adapter;
+	struct ixgbe_hw *hw;
 	u32 mfval, manc, reg;
 	int num_filters = 4;
 	bool manc_ipv4;
@@ -497,6 +500,12 @@ static int ixgbe_ipsec_check_mgmt_ip(struct xfrm_state *xs)
 #define BMCIP_V6                 0x3
 #define BMCIP_MASK               0x3
 
+	if (xs->xso.slave_dev)
+		dev = xs->xso.slave_dev;
+
+	adapter = netdev_priv(dev);
+	hw = &adapter->hw;
+
 	manc = IXGBE_READ_REG(hw, IXGBE_MANC);
 	manc_ipv4 = !!(manc & MANC_EN_IPV4_FILTER);
 	mfval = IXGBE_READ_REG(hw, IXGBE_MFVAL);
@@ -561,14 +570,21 @@ static int ixgbe_ipsec_check_mgmt_ip(struct xfrm_state *xs)
 static int ixgbe_ipsec_add_sa(struct xfrm_state *xs)
 {
 	struct net_device *dev = xs->xso.dev;
-	struct ixgbe_adapter *adapter = netdev_priv(dev);
-	struct ixgbe_ipsec *ipsec = adapter->ipsec;
-	struct ixgbe_hw *hw = &adapter->hw;
+	struct ixgbe_adapter *adapter;
+	struct ixgbe_ipsec *ipsec;
+	struct ixgbe_hw *hw;
 	int checked, match, first;
 	u16 sa_idx;
 	int ret;
 	int i;
 
+	if (xs->xso.slave_dev)
+		dev = xs->xso.slave_dev;
+
+	adapter = netdev_priv(dev);
+	ipsec = adapter->ipsec;
+	hw = &adapter->hw;
+
 	if (xs->id.proto != IPPROTO_ESP && xs->id.proto != IPPROTO_AH) {
 		netdev_err(dev, "Unsupported protocol 0x%04x for ipsec offload\n",
 			   xs->id.proto);
@@ -746,12 +762,19 @@ static int ixgbe_ipsec_add_sa(struct xfrm_state *xs)
 static void ixgbe_ipsec_del_sa(struct xfrm_state *xs)
 {
 	struct net_device *dev = xs->xso.dev;
-	struct ixgbe_adapter *adapter = netdev_priv(dev);
-	struct ixgbe_ipsec *ipsec = adapter->ipsec;
-	struct ixgbe_hw *hw = &adapter->hw;
+	struct ixgbe_adapter *adapter;
+	struct ixgbe_ipsec *ipsec;
+	struct ixgbe_hw *hw;
 	u32 zerobuf[4] = {0, 0, 0, 0};
 	u16 sa_idx;
 
+	if (xs->xso.slave_dev)
+		dev = xs->xso.slave_dev;
+
+	adapter = netdev_priv(dev);
+	ipsec = adapter->ipsec;
+	hw = &adapter->hw;
+
 	if (xs->xso.flags & XFRM_OFFLOAD_INBOUND) {
 		struct rx_sa *rsa;
 		u8 ipi;

drivers/net/ethernet/mellanox/mlx5/core/en_accel/ipsec.c

@@ -210,6 +210,9 @@ static inline int mlx5e_xfrm_validate_state(struct xfrm_state *x)
 	struct net_device *netdev = x->xso.dev;
 	struct mlx5e_priv *priv;
 
+	if (x->xso.slave_dev)
+		netdev = x->xso.slave_dev;
+
 	priv = netdev_priv(netdev);
 
 	if (x->props.aalgo != SADB_AALG_NONE) {
@@ -291,6 +294,9 @@ static int mlx5e_xfrm_add_state(struct xfrm_state *x)
 	unsigned int sa_handle;
 	int err;
 
+	if (x->xso.slave_dev)
+		netdev = x->xso.slave_dev;
+
 	priv = netdev_priv(netdev);
 
 	err = mlx5e_xfrm_validate_state(x);

include/net/bonding.h

@@ -238,6 +238,9 @@ struct bonding {
 	struct	 dentry *debug_dir;
 #endif /* CONFIG_DEBUG_FS */
 	struct rtnl_link_stats64 bond_stats;
+#ifdef CONFIG_XFRM_OFFLOAD
+	struct xfrm_state *xs;
+#endif /* CONFIG_XFRM_OFFLOAD */
 };
 
 #define bond_slave_get_rcu(dev) \

include/net/xfrm.h

@@ -127,6 +127,7 @@ struct xfrm_state_walk {
 
 struct xfrm_state_offload {
 	struct net_device	*dev;
+	struct net_device	*slave_dev;
 	unsigned long		offload_handle;
 	unsigned int		num_exthdrs;
 	u8			flags;

net/xfrm/xfrm_device.c

@@ -106,6 +106,7 @@ struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t featur
 	struct sk_buff *skb2, *nskb, *pskb = NULL;
 	netdev_features_t esp_features = features;
 	struct xfrm_offload *xo = xfrm_offload(skb);
+	struct net_device *dev = skb->dev;
 	struct sec_path *sp;
 
 	if (!xo)
@@ -119,6 +120,10 @@ struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t featur
 	if (xo->flags & XFRM_GRO || x->xso.flags & XFRM_OFFLOAD_INBOUND)
 		return skb;
 
+	/* This skb was already validated on the master dev */
+	if ((x->xso.dev != dev) && (x->xso.slave_dev == dev))
+		return skb;
+
 	local_irq_save(flags);
 	sd = this_cpu_ptr(&softnet_data);
 	err = !skb_queue_empty(&sd->xfrm_backlog);
@@ -129,15 +134,11 @@ struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t featur
 		return skb;
 	}
 
-	if (skb_is_gso(skb)) {
-		struct net_device *dev = skb->dev;
-
-		if (unlikely(x->xso.dev != dev)) {
+	if (skb_is_gso(skb) && unlikely(x->xso.dev != dev)) {
 		struct sk_buff *segs;
 
 		/* Packet got rerouted, fixup features and segment it. */
-			esp_features = esp_features & ~(NETIF_F_HW_ESP
-							| NETIF_F_GSO_ESP);
+		esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
 
 		segs = skb_gso_segment(skb, esp_features);
 		if (IS_ERR(segs)) {
@@ -149,7 +150,6 @@ struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t featur
 			skb = segs;
 		}
 	}
-	}
 
 	if (!skb->next) {
 		esp_features |= skb->dev->gso_partial_features;