[RFC PATCH net-next 0/3] bonding: support hardware crypto offload

[RFC PATCH net-next 0/3] bonding: support hardware crypto offload

[Jarod Wilson]

This is an initial "proof of concept" functional implementation for doing
pass-through of hardware encryption from bonding device to capable slaves.
This was tested using an ixgbe-driven Intel x520 NIC with libreswan and a
transport mode connection, on top of an active-backup bond, using netperf
and downing an interface during. Failover takes a moment, but does work,
and overall performance is right on par with offload when running on a
bare interface.

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 to update driver code accordingly.

I'm hoping folks with more of an idea about xfrm have some thoughts on
ways to make this cleaner, and possibly support more bonding modes, but
I'm reasonably happy I've made it this far. :)

Jarod Wilson (3):
  xfrm: bail early on slave pass over skb
  ixgbe_ipsec: become aware of when running as a bonding slave
  bonding: support hardware encryption offload to slaves

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org

 drivers/net/bonding/bond_main.c               | 103 +++++++++++++++++-
 .../net/ethernet/intel/ixgbe/ixgbe_ipsec.c    |  39 +++++--
 include/net/bonding.h                         |   1 +
 include/net/xfrm.h                            |   1 +
 net/xfrm/xfrm_device.c                        |  34 +++---
 5 files changed, 150 insertions(+), 28 deletions(-)

-- 
2.20.1

[RFC PATCH net-next 1/3] xfrm: bail early on slave pass over skb

[Jarod Wilson]

This is prep work for initial support of bonding hardware encryption
pass-through support. The bonding driver will fill in the slave_dev
pointer, and we use that to know not to skb_push() again on a given
skb that was already processed on the bond device.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
---
 include/net/xfrm.h     |  1 +
 net/xfrm/xfrm_device.c | 34 +++++++++++++++++-----------------
 2 files changed, 18 insertions(+), 17 deletions(-)

diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index 8f71c111e65a..a6ec341cd9f0 100644
--- a/include/net/xfrm.h
+++ b/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;
diff --git a/net/xfrm/xfrm_device.c b/net/xfrm/xfrm_device.c
index 6cc7f7f1dd68..1cd31dcf59da 100644
--- a/net/xfrm/xfrm_device.c
+++ b/net/xfrm/xfrm_device.c
@@ -108,6 +108,7 @@ 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)
@@ -121,6 +122,10 @@ 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);
@@ -131,25 +136,20 @@ 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)) {
-			struct sk_buff *segs;
+	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);
+		/* Packet got rerouted, fixup features and segment it. */
+		esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
 
-			segs = skb_gso_segment(skb, esp_features);
-			if (IS_ERR(segs)) {
-				kfree_skb(skb);
-				atomic_long_inc(&dev->tx_dropped);
-				return NULL;
-			} else {
-				consume_skb(skb);
-				skb = segs;
-			}
+		segs = skb_gso_segment(skb, esp_features);
+		if (IS_ERR(segs)) {
+			kfree_skb(skb);
+			atomic_long_inc(&dev->tx_dropped);
+			return NULL;
+		} else {
+			consume_skb(skb);
+			skb = segs;
 		}
 	}
 
-- 
2.20.1

[RFC PATCH net-next 2/3] ixgbe_ipsec: become aware of when running as a bonding slave

[Jarod Wilson]

Slave devices in a bond doing hardware encryption also need to be aware
that they're slaves, so we operate on the slave instead of the bonding
master to do the actual hardware encryption offload bits.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
---
 .../net/ethernet/intel/ixgbe/ixgbe_ipsec.c    | 39 +++++++++++++++----
 1 file changed, 31 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_ipsec.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_ipsec.c
index 113f6087c7c9..26b0a58a064d 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_ipsec.c
+++ b/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;
-- 
2.20.1

[RFC PATCH net-next 3/3] bonding: support hardware encryption offload to slaves

[Jarod Wilson]

Currently, this support is limited to active-backup mode, as I'm not sure
about the feasilibity of mapping an xfrm_state's offload handle to
multiple hardware devices simultaneously, and we rely on being able to
pass some hints to both the xfrm and NIC driver about whether or not
they're operating on a slave device.

I've tested this atop an Intel x520 device (ixgbe) using libreswan in
transport mode, succesfully achieving ~4.3Gbps throughput with netperf
(more or less identical to throughput on a bare NIC in this system),
as well as successful failover and recovery mid-netperf.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
---
 drivers/net/bonding/bond_main.c | 103 +++++++++++++++++++++++++++++++-
 include/net/bonding.h           |   1 +
 2 files changed, 101 insertions(+), 3 deletions(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index 2e70e43c5df5..781da5beb484 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/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 @@ void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 		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 +373,82 @@ static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 	return 0;
 }
 
+/*---------------------------------- XFRM -----------------------------------*/
+
+/**
+ * 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,
+};
+
 /*------------------------------- Link status -------------------------------*/
 
 /* Set the carrier state for the master according to the state of its
@@ -878,6 +955,9 @@ void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 	if (old_active == new_active)
 		return;
 
+	if ((BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) && bond->xs)
+		bond_ipsec_del_sa(bond->xs);
+
 	if (new_active) {
 		new_active->last_link_up = jiffies;
 
@@ -941,6 +1021,11 @@ void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 					bond_should_notify_peers(bond);
 			}
 
+			if (old_active && bond->xs) {
+				xfrm_dev_state_flush(dev_net(bond->dev), bond->dev, true);
+				bond_ipsec_add_sa(bond->xs);
+			}
+
 			call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 			if (should_notify_peers) {
 				bond->send_peer_notif--;
@@ -1125,7 +1210,9 @@ static netdev_features_t bond_fix_features(struct net_device *dev,
 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
 
 #define BOND_ENC_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
-				 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
+				 NETIF_F_RXCSUM | NETIF_F_ALL_TSO | \
+				 NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM | \
+				 NETIF_F_GSO_ESP)
 
 #define BOND_MPLS_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
 				 NETIF_F_ALL_TSO)
@@ -1464,6 +1551,9 @@ 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
@@ -4444,6 +4534,11 @@ 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);
 
+	/* 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;
+
 	/* don't acquire bond device's netif_tx_lock when transmitting */
 	bond_dev->features |= NETIF_F_LLTX;
 
@@ -4462,6 +4557,8 @@ 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;
+	if ((BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP))
+		bond_dev->hw_features |= BOND_ENC_FEATURES;
 	bond_dev->features |= bond_dev->hw_features;
 	bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
 }
diff --git a/include/net/bonding.h b/include/net/bonding.h
index dc2ce31a1f52..854b46511bd8 100644
--- a/include/net/bonding.h
+++ b/include/net/bonding.h
@@ -238,6 +238,7 @@ struct bonding {
 #endif /* CONFIG_DEBUG_FS */
 	struct rtnl_link_stats64 bond_stats;
 	struct lock_class_key stats_lock_key;
+	struct xfrm_state *xs;
 };
 
 #define bond_slave_get_rcu(dev) \
-- 
2.20.1

[RFC PATCH net-next v2 0/3] bonding: support hardware crypto offload

[Jarod Wilson]

This is an initial "proof of concept" functional implementation for doing
pass-through of hardware encryption from bonding device to capable slaves.
This was tested using an ixgbe-driven Intel x520 NIC with libreswan and a
transport mode connection, on top of an active-backup bond, using netperf
and downing an interface during. Failover takes a moment, but does work,
and overall performance is right on par with offload when running on a
bare interface.

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 to update driver code accordingly.

I'm hoping folks with more of an idea about xfrm have some thoughts on
ways to make this cleaner, and possibly support more bonding modes, but
I'm reasonably happy I've made it this far. :)

v2: fix build with CONFIG_XFRM_OFFLOAD disabled and rebase on latest
    net-next tree bonding changes

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org

Jarod Wilson (3):
  xfrm: bail early on slave pass over skb
  ixgbe_ipsec: become aware of when running as a bonding slave
  bonding: support hardware encryption offload to slaves

 drivers/net/bonding/bond_main.c               | 111 +++++++++++++++++-
 .../net/ethernet/intel/ixgbe/ixgbe_ipsec.c    |  39 ++++--
 include/net/bonding.h                         |   3 +
 include/net/xfrm.h                            |   1 +
 net/xfrm/xfrm_device.c                        |  34 +++---
 5 files changed, 160 insertions(+), 28 deletions(-)

-- 
2.20.1

[RFC PATCH net-next v2 1/3] xfrm: bail early on slave pass over skb

[Jarod Wilson]

This is prep work for initial support of bonding hardware encryption
pass-through support. The bonding driver will fill in the slave_dev
pointer, and we use that to know not to skb_push() again on a given
skb that was already processed on the bond device.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
---
 include/net/xfrm.h     |  1 +
 net/xfrm/xfrm_device.c | 34 +++++++++++++++++-----------------
 2 files changed, 18 insertions(+), 17 deletions(-)

diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index 8f71c111e65a..a6ec341cd9f0 100644
--- a/include/net/xfrm.h
+++ b/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;
diff --git a/net/xfrm/xfrm_device.c b/net/xfrm/xfrm_device.c
index 6cc7f7f1dd68..1cd31dcf59da 100644
--- a/net/xfrm/xfrm_device.c
+++ b/net/xfrm/xfrm_device.c
@@ -108,6 +108,7 @@ 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)
@@ -121,6 +122,10 @@ 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);
@@ -131,25 +136,20 @@ 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)) {
-			struct sk_buff *segs;
+	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);
+		/* Packet got rerouted, fixup features and segment it. */
+		esp_features = esp_features & ~(NETIF_F_HW_ESP | NETIF_F_GSO_ESP);
 
-			segs = skb_gso_segment(skb, esp_features);
-			if (IS_ERR(segs)) {
-				kfree_skb(skb);
-				atomic_long_inc(&dev->tx_dropped);
-				return NULL;
-			} else {
-				consume_skb(skb);
-				skb = segs;
-			}
+		segs = skb_gso_segment(skb, esp_features);
+		if (IS_ERR(segs)) {
+			kfree_skb(skb);
+			atomic_long_inc(&dev->tx_dropped);
+			return NULL;
+		} else {
+			consume_skb(skb);
+			skb = segs;
 		}
 	}
 
-- 
2.20.1

[RFC PATCH net-next v2 2/3] ixgbe_ipsec: become aware of when running as a bonding slave

[Jarod Wilson]

Slave devices in a bond doing hardware encryption also need to be aware
that they're slaves, so we operate on the slave instead of the bonding
master to do the actual hardware encryption offload bits.

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>
---
 .../net/ethernet/intel/ixgbe/ixgbe_ipsec.c    | 39 +++++++++++++++----
 1 file changed, 31 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/intel/ixgbe/ixgbe_ipsec.c b/drivers/net/ethernet/intel/ixgbe/ixgbe_ipsec.c
index 113f6087c7c9..26b0a58a064d 100644
--- a/drivers/net/ethernet/intel/ixgbe/ixgbe_ipsec.c
+++ b/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;
-- 
2.20.1

[RFC PATCH net-next v2 3/3] bonding: support hardware encryption offload to slaves

[Jarod Wilson]

Currently, this support is limited to active-backup mode, as I'm not sure
about the feasilibity of mapping an xfrm_state's offload handle to
multiple hardware devices simultaneously, and we rely on being able to
pass some hints to both the xfrm and NIC driver about whether or not
they're operating on a slave device.

I've tested this atop an Intel x520 device (ixgbe) using libreswan in
transport mode, succesfully achieving ~4.3Gbps throughput with netperf
(more or less identical to throughput on a bare NIC in this system),
as well as successful failover and recovery mid-netperf.

v2: rebase on latest net-next and wrap with #ifdef CONFIG_XFRM_OFFLOAD

CC: Jay Vosburgh <j.vosburgh@gmail.com>
CC: Veaceslav Falico <vfalico@gmail.com>
CC: Andy Gospodarek <andy@greyhouse.net>
CC: "David S. Miller" <davem@davemloft.net>
CC: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
CC: Jakub Kicinski <kuba@kernel.org>
CC: Steffen Klassert <steffen.klassert@secunet.com>
CC: Herbert Xu <herbert@gondor.apana.org.au>
CC: netdev@vger.kernel.org
CC: intel-wired-lan@lists.osuosl.org
Signed-off-by: Jarod Wilson <jarod@redhat.com>

Signed-off-by: Jarod Wilson <jarod@redhat.com>
---
 drivers/net/bonding/bond_main.c | 111 +++++++++++++++++++++++++++++++-
 include/net/bonding.h           |   3 +
 2 files changed, 111 insertions(+), 3 deletions(-)

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index baa93191dfdd..b90a86029df5 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/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 @@ void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 		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
@@ -878,6 +957,11 @@ void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 	if (old_active == new_active)
 		return;
 
+#ifdef CONFIG_XFRM_OFFLOAD
+	if ((BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) && bond->xs)
+		bond_ipsec_del_sa(bond->xs);
+#endif /* CONFIG_XFRM_OFFLOAD */
+
 	if (new_active) {
 		new_active->last_link_up = jiffies;
 
@@ -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--;
@@ -1125,7 +1216,9 @@ static netdev_features_t bond_fix_features(struct net_device *dev,
 				 NETIF_F_HIGHDMA | NETIF_F_LRO)
 
 #define BOND_ENC_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
-				 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
+				 NETIF_F_RXCSUM | NETIF_F_ALL_TSO | \
+				 NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM | \
+				 NETIF_F_GSO_ESP)
 
 #define BOND_MPLS_FEATURES	(NETIF_F_HW_CSUM | NETIF_F_SG | \
 				 NETIF_F_ALL_TSO)
@@ -1464,6 +1557,9 @@ 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
@@ -4444,6 +4540,13 @@ 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;
 
@@ -4462,6 +4565,8 @@ 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;
+	if ((BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP))
+		bond_dev->hw_features |= BOND_ENC_FEATURES;
 	bond_dev->features |= bond_dev->hw_features;
 	bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
 }
diff --git a/include/net/bonding.h b/include/net/bonding.h
index 0b696da5c115..4ca178b2d0e2 100644
--- a/include/net/bonding.h
+++ b/include/net/bonding.h
@@ -237,6 +237,9 @@ 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) \
-- 
2.20.1