diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 6c7960c8..8e9f8cb 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -110,6 +110,7 @@ enum netdev_tx {
 	__NETDEV_TX_MIN	 = INT_MIN,	/* make sure enum is signed */
 	NETDEV_TX_OK	 = 0x00,	/* driver took care of packet */
 	NETDEV_TX_BUSY	 = 0x10,	/* driver tx path was busy*/
+	NETDEV_TX_AGAIN	 = 0x20,	/* tx path needs resubmission */
 };
 typedef enum netdev_tx netdev_tx_t;
 
@@ -2771,7 +2772,9 @@ struct softnet_data {
 	struct Qdisc		*output_queue;
 	struct Qdisc		**output_queue_tailp;
 	struct sk_buff		*completion_queue;
-
+#ifdef CONFIG_XFRM_OFFLOAD
+	struct sk_buff_head	xfrm_backlog;
+#endif
 #ifdef CONFIG_RPS
 	/* input_queue_head should be written by cpu owning this struct,
 	 * and only read by other cpus. Worth using a cache line.
@@ -3303,7 +3306,7 @@ int dev_get_phys_port_id(struct net_device *dev,
 int dev_get_phys_port_name(struct net_device *dev,
 			   char *name, size_t len);
 int dev_change_proto_down(struct net_device *dev, bool proto_down);
-struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev);
+struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, int *ret);
 struct sk_buff *dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
 				    struct netdev_queue *txq, int *ret);
 
diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index 08efc16..919fc97 100644
--- a/include/net/xfrm.h
+++ b/include/net/xfrm.h
@@ -1020,6 +1020,7 @@ struct xfrm_offload {
 #define	XFRM_GSO_SEGMENT	16
 #define	XFRM_GRO		32
 #define	XFRM_ESP_NO_TRAILER	64
+#define	XFRM_DEV_RESUME		128
 
 	__u32			status;
 #define CRYPTO_SUCCESS				1
@@ -1857,7 +1858,9 @@ static inline struct xfrm_offload *xfrm_offload(struct sk_buff *skb)
 void __net_init xfrm_dev_init(void);
 
 #ifdef CONFIG_XFRM_OFFLOAD
-struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features);
+void xfrm_dev_resume(struct sk_buff *skb);
+void xfrm_dev_backlog(struct softnet_data *sd);
+struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, int *ret);
 int xfrm_dev_state_add(struct net *net, struct xfrm_state *x,
 		       struct xfrm_user_offload *xuo);
 bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x);
@@ -1896,7 +1899,15 @@ static inline void xfrm_dev_state_free(struct xfrm_state *x)
 	}
 }
 #else
-static inline struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features)
+static inline void xfrm_dev_resume(struct sk_buff *skb)
+{
+}
+
+static inline void xfrm_dev_backlog(struct softnet_data *sd)
+{
+}
+
+static inline struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, int *ret)
 {
 	return skb;
 }
diff --git a/net/core/dev.c b/net/core/dev.c
index d278cf9..7cc4a3b0 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3059,7 +3059,7 @@ int skb_csum_hwoffload_help(struct sk_buff *skb,
 }
 EXPORT_SYMBOL(skb_csum_hwoffload_help);
 
-static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
+static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, int *ret)
 {
 	netdev_features_t features;
 
@@ -3099,7 +3099,7 @@ static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device
 		}
 	}
 
-	skb = validate_xmit_xfrm(skb, features);
+	skb = validate_xmit_xfrm(skb, features, ret);
 
 	return skb;
 
@@ -3110,7 +3110,7 @@ static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device
 	return NULL;
 }
 
-struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
+struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, int *ret)
 {
 	struct sk_buff *next, *head = NULL, *tail;
 
@@ -3121,7 +3121,7 @@ struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *d
 		/* in case skb wont be segmented, point to itself */
 		skb->prev = skb;
 
-		skb = validate_xmit_skb(skb, dev);
+		skb = validate_xmit_skb(skb, dev, ret);
 		if (!skb)
 			continue;
 
@@ -3493,7 +3493,7 @@ static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
 				     XMIT_RECURSION_LIMIT))
 				goto recursion_alert;
 
-			skb = validate_xmit_skb(skb, dev);
+			skb = validate_xmit_skb(skb, dev, &rc);
 			if (!skb)
 				goto out;
 
@@ -4174,6 +4174,8 @@ static __latent_entropy void net_tx_action(struct softirq_action *h)
 			spin_unlock(root_lock);
 		}
 	}
+
+	xfrm_dev_backlog(sd);
 }
 
 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
@@ -8812,6 +8814,9 @@ static int __init net_dev_init(void)
 
 		skb_queue_head_init(&sd->input_pkt_queue);
 		skb_queue_head_init(&sd->process_queue);
+#ifdef CONFIG_XFRM_OFFLOAD
+		skb_queue_head_init(&sd->xfrm_backlog);
+#endif
 		INIT_LIST_HEAD(&sd->poll_list);
 		sd->output_queue_tailp = &sd->output_queue;
 #ifdef CONFIG_RPS
diff --git a/net/ipv4/esp4.c b/net/ipv4/esp4.c
index b00e4a4..d61dcdc 100644
--- a/net/ipv4/esp4.c
+++ b/net/ipv4/esp4.c
@@ -121,14 +121,32 @@ static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
 static void esp_output_done(struct crypto_async_request *base, int err)
 {
 	struct sk_buff *skb = base->data;
+	struct xfrm_offload *xo = xfrm_offload(skb);
 	void *tmp;
-	struct dst_entry *dst = skb_dst(skb);
-	struct xfrm_state *x = dst->xfrm;
+	struct xfrm_state *x;
+
+	if (xo && (xo->flags & XFRM_DEV_RESUME))
+		x = skb->sp->xvec[skb->sp->len - 1];
+	else
+		x = skb_dst(skb)->xfrm;
 
 	tmp = ESP_SKB_CB(skb)->tmp;
 	esp_ssg_unref(x, tmp);
 	kfree(tmp);
-	xfrm_output_resume(skb, err);
+
+	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
+		if (err) {
+			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
+			kfree_skb(skb);
+			return;
+		}
+
+		skb_push(skb, skb->data - skb_mac_header(skb));
+		secpath_reset(skb);
+		xfrm_dev_resume(skb);
+	} else {
+		xfrm_output_resume(skb, err);
+	}
 }
 
 /* Move ESP header back into place. */
diff --git a/net/ipv6/esp6.c b/net/ipv6/esp6.c
index 89910e2..1c071c3 100644
--- a/net/ipv6/esp6.c
+++ b/net/ipv6/esp6.c
@@ -141,14 +141,32 @@ static void esp_ssg_unref(struct xfrm_state *x, void *tmp)
 static void esp_output_done(struct crypto_async_request *base, int err)
 {
 	struct sk_buff *skb = base->data;
+	struct xfrm_offload *xo = xfrm_offload(skb);
 	void *tmp;
-	struct dst_entry *dst = skb_dst(skb);
-	struct xfrm_state *x = dst->xfrm;
+	struct xfrm_state *x;
+
+	if (xo && (xo->flags & XFRM_DEV_RESUME))
+		x = skb->sp->xvec[skb->sp->len - 1];
+	else
+		x = skb_dst(skb)->xfrm;
 
 	tmp = ESP_SKB_CB(skb)->tmp;
 	esp_ssg_unref(x, tmp);
 	kfree(tmp);
-	xfrm_output_resume(skb, err);
+
+	if (xo && (xo->flags & XFRM_DEV_RESUME)) {
+		if (err) {
+			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
+			kfree_skb(skb);
+			return;
+		}
+
+		skb_push(skb, skb->data - skb_mac_header(skb));
+		secpath_reset(skb);
+		xfrm_dev_resume(skb);
+	} else {
+		xfrm_output_resume(skb, err);
+	}
 }
 
 /* Move ESP header back into place. */
diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index 4f4fa32..e536ed3 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -255,7 +255,7 @@ static int packet_direct_xmit(struct sk_buff *skb)
 		     !netif_carrier_ok(dev)))
 		goto drop;
 
-	skb = validate_xmit_skb_list(skb, dev);
+	skb = validate_xmit_skb_list(skb, dev, &ret);
 	if (skb != orig_skb)
 		goto drop;
 
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 6ced7c8..cdbbd30 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -30,6 +30,7 @@
 #include <net/pkt_sched.h>
 #include <net/dst.h>
 #include <trace/events/qdisc.h>
+#include <net/xfrm.h>
 
 /* Qdisc to use by default */
 const struct Qdisc_ops *default_qdisc_ops = &pfifo_fast_ops;
@@ -119,6 +120,8 @@ static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
 	if (unlikely(skb)) {
 		/* skb in gso_skb were already validated */
 		*validate = false;
+		if (xfrm_offload(skb))
+			*validate = true;
 		/* check the reason of requeuing without tx lock first */
 		txq = skb_get_tx_queue(txq->dev, skb);
 		if (!netif_xmit_frozen_or_stopped(txq)) {
@@ -178,7 +181,19 @@ int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
 
 	/* Note that we validate skb (GSO, checksum, ...) outside of locks */
 	if (validate)
-		skb = validate_xmit_skb_list(skb, dev);
+		skb = validate_xmit_skb_list(skb, dev, &ret);
+
+#ifdef CONFIG_XFRM_OFFLOAD
+	if (unlikely(ret == NETDEV_TX_AGAIN)) {
+		spin_lock(root_lock);
+		ret = dev_requeue_skb(skb, q);
+
+		if (ret && netif_xmit_frozen_or_stopped(txq))
+			ret = 0;
+
+		return ret;
+	}
+#endif
 
 	if (likely(skb)) {
 		HARD_TX_LOCK(dev, txq, smp_processor_id());
diff --git a/net/xfrm/xfrm_device.c b/net/xfrm/xfrm_device.c
index 5c70d2d..1e0fed6 100644
--- a/net/xfrm/xfrm_device.c
+++ b/net/xfrm/xfrm_device.c
@@ -23,12 +23,13 @@
 #include <linux/notifier.h>
 
 #ifdef CONFIG_XFRM_OFFLOAD
-struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features)
+struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features, int *ret)
 {
 	int err;
-	__u32 seq;
+	unsigned long flags;
 	struct xfrm_state *x;
 	struct sk_buff *skb2;
+	struct softnet_data *sd;
 	netdev_features_t esp_features = features;
 	struct xfrm_offload *xo = xfrm_offload(skb);
 
@@ -42,6 +43,21 @@ 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;
 
+	err = 0;
+
+	*ret = NETDEV_TX_OK;
+
+	local_irq_save(flags);
+	sd = this_cpu_ptr(&softnet_data);
+	if (!skb_queue_empty(&sd->xfrm_backlog))
+		err = -EBUSY;
+	local_irq_restore(flags);
+
+	if (err) {
+		*ret = NETDEV_TX_AGAIN;
+		return skb;
+	}
+
 	if (skb_is_gso(skb)) {
 		struct net_device *dev = skb->dev;
 
@@ -54,23 +70,26 @@ struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t featur
 
 			segs = skb_gso_segment(skb, esp_features);
 			if (IS_ERR(segs)) {
-				XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 				kfree_skb(skb);
+				atomic_long_inc(&dev->tx_dropped);
 				return NULL;
 			} else {
 				consume_skb(skb);
 				skb = segs;
 			}
-		} else {
-			return skb;
 		}
 	}
 
 	if (!skb->next) {
 		x->outer_mode->xmit(x, skb);
 
+		xo->flags |= XFRM_DEV_RESUME;
+
 		err = x->type_offload->xmit(x, skb, esp_features);
 		if (err) {
+			if (err == -EINPROGRESS)
+				return NULL;
+
 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 			kfree_skb(skb);
 			return NULL;
@@ -82,36 +101,37 @@ struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t featur
 	}
 
 	skb2 = skb;
-	seq = xo->seq.low;
 
 	do {
 		struct sk_buff *nskb = skb2->next;
+		skb2->next = NULL;
 
 		xo = xfrm_offload(skb2);
-		xo->flags |= XFRM_GSO_SEGMENT;
-		xo->seq.low = seq;
-		xo->seq.hi = xfrm_replay_seqhi(x, seq);
-
-		if(!(features & NETIF_F_HW_ESP))
-			xo->flags |= CRYPTO_FALLBACK;
+		xo->flags |= XFRM_DEV_RESUME;
 
 		x->outer_mode->xmit(x, skb2);
 
 		err = x->type_offload->xmit(x, skb2, esp_features);
-		if (err) {
+		if (!err) {
+			skb2->next = nskb;
+		} else if (err != -EINPROGRESS) {
 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
 			skb2->next = nskb;
 			kfree_skb_list(skb2);
 			return NULL;
-		}
+		} else {
+			if (skb == skb2)
+				skb = nskb;
 
-		if (!skb_is_gso(skb2))
-			seq++;
-		else
-			seq += skb_shinfo(skb2)->gso_segs;
+			if (!skb)
+				return NULL;
+
+			goto skip_push;
+		}
 
 		skb_push(skb2, skb2->data - skb_mac_header(skb2));
 
+skip_push:
 		skb2 = nskb;
 	} while (skb2);
 
@@ -207,6 +227,55 @@ bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
 	return true;
 }
 EXPORT_SYMBOL_GPL(xfrm_dev_offload_ok);
+
+void xfrm_dev_resume(struct sk_buff *skb)
+{
+	int ret = NETDEV_TX_BUSY;
+	unsigned long flags;
+	struct netdev_queue *txq;
+	struct softnet_data *sd;
+	struct net_device *dev = skb->dev;
+
+	rcu_read_lock();
+	txq = netdev_pick_tx(dev, skb, NULL);
+
+	HARD_TX_LOCK(dev, txq, smp_processor_id());
+	if (!netif_xmit_frozen_or_stopped(txq))
+		skb = dev_hard_start_xmit(skb, dev, txq, &ret);
+	HARD_TX_UNLOCK(dev, txq);
+
+	if (!dev_xmit_complete(ret)) {
+		local_irq_save(flags);
+		sd = this_cpu_ptr(&softnet_data);
+		skb_queue_tail(&sd->xfrm_backlog, skb);
+		raise_softirq_irqoff(NET_TX_SOFTIRQ);
+		local_irq_restore(flags);
+	}
+	rcu_read_unlock();
+}
+EXPORT_SYMBOL_GPL(xfrm_dev_resume);
+
+void xfrm_dev_backlog(struct softnet_data *sd)
+{
+	struct sk_buff *skb;
+	struct sk_buff_head list;
+	struct sk_buff_head *xfrm_backlog = &sd->xfrm_backlog;
+
+	if (skb_queue_empty(xfrm_backlog))
+		return;
+
+	__skb_queue_head_init(&list);
+
+	spin_lock(&xfrm_backlog->lock);
+	skb_queue_splice_init(xfrm_backlog, &list);
+	spin_unlock(&xfrm_backlog->lock);
+
+	while (!skb_queue_empty(&list)) {
+		skb = __skb_dequeue(&list);
+		xfrm_dev_resume(skb);
+	}
+
+}
 #endif
 
 static int xfrm_dev_register(struct net_device *dev)