[PATCH RFC v2 0/5] Support asynchronous crypto for IPsec GSO.

[PATCH RFC v2 0/5] Support asynchronous crypto for IPsec GSO.

[Steffen Klassert]

This patchset implements asynchronous crypto handling
in the layer 2 TX path. With this we can allow IPsec
ESP GSO for software crypto. This also merges the IPsec
GSO and non-GSO paths to both use validate_xmit_xfrm().

1) Separate ESP handling from segmentation for GRO packets.
   This unifies the IPsec GSO and non GSO codepath.

2) Add asynchronous callbacks for xfrm on layer 2. This
   adds the necessary infrastructure to core networking.

3) Allow to use the layer2 IPsec GSO codepath for software
   crypto, all infrastructure is there now.

4) Also allow IPsec GSO with software crypto for local sockets.

5) Don't require synchronous crypto fallback on IPsec offloading,
   it is not needed anymore.

I plan to merge this to ipsec-next during this development cycle.

Changes from v1:

- Resolve conflicts with the lockless qdisc patchset.

- Fix compilation error when CONFIG_XFRM is not set.

- Order local local variables of new functions in reverse christmas
  tree ordering.

[PATCH RFC v2 1/5] xfrm: Separate ESP handling from segmentation for GRO packets.

[Steffen Klassert]

We change the ESP GSO handlers to only segment the packets.
The ESP handling and encryption is defered to validate_xmit_xfrm()
where this is done for non GRO packets too. This makes the code
more robust and prepares for asynchronous crypto handling.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/net/xfrm.h           |  6 +--
 net/core/dev.c               |  5 +--
 net/ipv4/esp4_offload.c      | 73 +++++++++++--------------------------
 net/ipv4/xfrm4_mode_tunnel.c |  5 +--
 net/ipv6/esp6_offload.c      | 80 ++++++++++++----------------------------
 net/ipv6/xfrm6_mode_tunnel.c |  5 +--
 net/xfrm/xfrm_device.c       | 87 +++++++++++++++++++++++++++++++++++++++-----
 7 files changed, 129 insertions(+), 132 deletions(-)

diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index 1ec0c4760646..df7f3d0ac4a1 100644
--- a/include/net/xfrm.h
+++ b/include/net/xfrm.h
@@ -1888,7 +1888,7 @@ static inline struct xfrm_offload *xfrm_offload(struct sk_buff *skb)
 void __net_init xfrm_dev_init(void);
 
 #ifdef CONFIG_XFRM_OFFLOAD
-int 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 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);
@@ -1929,9 +1929,9 @@ static inline void xfrm_dev_state_free(struct xfrm_state *x)
 	}
 }
 #else
-static inline int validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features)
+static inline struct sk_buff *validate_xmit_xfrm(struct sk_buff *skb, netdev_features_t features)
 {
-	return 0;
+	return skb;
 }
 
 static inline int xfrm_dev_state_add(struct net *net, struct xfrm_state *x, struct xfrm_user_offload *xuo)
diff --git a/net/core/dev.c b/net/core/dev.c
index b0eee49a2489..7b89cbd9ef4c 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3066,9 +3066,6 @@ static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device
 		    __skb_linearize(skb))
 			goto out_kfree_skb;
 
-		if (validate_xmit_xfrm(skb, features))
-			goto out_kfree_skb;
-
 		/* If packet is not checksummed and device does not
 		 * support checksumming for this protocol, complete
 		 * checksumming here.
@@ -3085,6 +3082,8 @@ static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device
 		}
 	}
 
+	skb = validate_xmit_xfrm(skb, features);
+
 	return skb;
 
 out_kfree_skb:
diff --git a/net/ipv4/esp4_offload.c b/net/ipv4/esp4_offload.c
index f8b918c766b0..c359f3cfeec3 100644
--- a/net/ipv4/esp4_offload.c
+++ b/net/ipv4/esp4_offload.c
@@ -108,75 +108,36 @@ static void esp4_gso_encap(struct xfrm_state *x, struct sk_buff *skb)
 static struct sk_buff *esp4_gso_segment(struct sk_buff *skb,
 				        netdev_features_t features)
 {
-	__u32 seq;
-	int err = 0;
-	struct sk_buff *skb2;
 	struct xfrm_state *x;
 	struct ip_esp_hdr *esph;
 	struct crypto_aead *aead;
-	struct sk_buff *segs = ERR_PTR(-EINVAL);
 	netdev_features_t esp_features = features;
 	struct xfrm_offload *xo = xfrm_offload(skb);
 
 	if (!xo)
-		goto out;
-
-	seq = xo->seq.low;
+		return ERR_PTR(-EINVAL);
 
 	x = skb->sp->xvec[skb->sp->len - 1];
 	aead = x->data;
 	esph = ip_esp_hdr(skb);
 
 	if (esph->spi != x->id.spi)
-		goto out;
+		return ERR_PTR(-EINVAL);
 
 	if (!pskb_may_pull(skb, sizeof(*esph) + crypto_aead_ivsize(aead)))
-		goto out;
+		return ERR_PTR(-EINVAL);
 
 	__skb_pull(skb, sizeof(*esph) + crypto_aead_ivsize(aead));
 
 	skb->encap_hdr_csum = 1;
 
-	if (!(features & NETIF_F_HW_ESP))
+	if (!(features & NETIF_F_HW_ESP) || !x->xso.offload_handle ||
+	    (x->xso.dev != skb->dev))
 		esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
 
-	segs = x->outer_mode->gso_segment(x, skb, esp_features);
-	if (IS_ERR_OR_NULL(segs))
-		goto out;
-
-	__skb_pull(skb, skb->data - skb_mac_header(skb));
-
-	skb2 = segs;
-	do {
-		struct sk_buff *nskb = skb2->next;
-
-		xo = xfrm_offload(skb2);
-		xo->flags |= XFRM_GSO_SEGMENT;
-		xo->seq.low = seq;
-		xo->seq.hi = xfrm_replay_seqhi(x, seq);
+	xo->flags |= XFRM_GSO_SEGMENT;
 
-		if(!(features & NETIF_F_HW_ESP))
-			xo->flags |= CRYPTO_FALLBACK;
-
-		x->outer_mode->xmit(x, skb2);
-
-		err = x->type_offload->xmit(x, skb2, esp_features);
-		if (err) {
-			kfree_skb_list(segs);
-			return ERR_PTR(err);
-		}
-
-		if (!skb_is_gso(skb2))
-			seq++;
-		else
-			seq += skb_shinfo(skb2)->gso_segs;
-
-		skb_push(skb2, skb2->mac_len);
-		skb2 = nskb;
-	} while (skb2);
-
-out:
-	return segs;
+	return x->outer_mode->gso_segment(x, skb, esp_features);
 }
 
 static int esp_input_tail(struct xfrm_state *x, struct sk_buff *skb)
@@ -203,6 +164,7 @@ static int esp_xmit(struct xfrm_state *x, struct sk_buff *skb,  netdev_features_
 	struct crypto_aead *aead;
 	struct esp_info esp;
 	bool hw_offload = true;
+	__u32 seq;
 
 	esp.inplace = true;
 
@@ -241,23 +203,30 @@ static int esp_xmit(struct xfrm_state *x, struct sk_buff *skb,  netdev_features_
 			return esp.nfrags;
 	}
 
+	seq = xo->seq.low;
+
 	esph = esp.esph;
 	esph->spi = x->id.spi;
 
 	skb_push(skb, -skb_network_offset(skb));
 
 	if (xo->flags & XFRM_GSO_SEGMENT) {
-		esph->seq_no = htonl(xo->seq.low);
-	} else {
-		ip_hdr(skb)->tot_len = htons(skb->len);
-		ip_send_check(ip_hdr(skb));
+		esph->seq_no = htonl(seq);
+
+		if (!skb_is_gso(skb))
+			xo->seq.low++;
+		else
+			xo->seq.low += skb_shinfo(skb)->gso_segs;
 	}
 
+	esp.seqno = cpu_to_be64(seq + ((u64)xo->seq.hi << 32));
+
+	ip_hdr(skb)->tot_len = htons(skb->len);
+	ip_send_check(ip_hdr(skb));
+
 	if (hw_offload)
 		return 0;
 
-	esp.seqno = cpu_to_be64(xo->seq.low + ((u64)xo->seq.hi << 32));
-
 	err = esp_output_tail(x, skb, &esp);
 	if (err)
 		return err;
diff --git a/net/ipv4/xfrm4_mode_tunnel.c b/net/ipv4/xfrm4_mode_tunnel.c
index 7d885a44dc9d..8affc6d83d58 100644
--- a/net/ipv4/xfrm4_mode_tunnel.c
+++ b/net/ipv4/xfrm4_mode_tunnel.c
@@ -105,18 +105,15 @@ static struct sk_buff *xfrm4_mode_tunnel_gso_segment(struct xfrm_state *x,
 {
 	__skb_push(skb, skb->mac_len);
 	return skb_mac_gso_segment(skb, features);
-
 }
 
 static void xfrm4_mode_tunnel_xmit(struct xfrm_state *x, struct sk_buff *skb)
 {
 	struct xfrm_offload *xo = xfrm_offload(skb);
 
-	if (xo->flags & XFRM_GSO_SEGMENT) {
-		skb->network_header = skb->network_header - x->props.header_len;
+	if (xo->flags & XFRM_GSO_SEGMENT)
 		skb->transport_header = skb->network_header +
 					sizeof(struct iphdr);
-	}
 
 	skb_reset_mac_len(skb);
 	pskb_pull(skb, skb->mac_len + x->props.header_len);
diff --git a/net/ipv6/esp6_offload.c b/net/ipv6/esp6_offload.c
index 333a478aa161..0bb7d54cf2cb 100644
--- a/net/ipv6/esp6_offload.c
+++ b/net/ipv6/esp6_offload.c
@@ -135,75 +135,36 @@ static void esp6_gso_encap(struct xfrm_state *x, struct sk_buff *skb)
 static struct sk_buff *esp6_gso_segment(struct sk_buff *skb,
 				        netdev_features_t features)
 {
-	__u32 seq;
-	int err = 0;
-	struct sk_buff *skb2;
 	struct xfrm_state *x;
 	struct ip_esp_hdr *esph;
 	struct crypto_aead *aead;
-	struct sk_buff *segs = ERR_PTR(-EINVAL);
 	netdev_features_t esp_features = features;
 	struct xfrm_offload *xo = xfrm_offload(skb);
 
 	if (!xo)
-		goto out;
-
-	seq = xo->seq.low;
+		return ERR_PTR(-EINVAL);
 
 	x = skb->sp->xvec[skb->sp->len - 1];
 	aead = x->data;
 	esph = ip_esp_hdr(skb);
 
 	if (esph->spi != x->id.spi)
-		goto out;
+		return ERR_PTR(-EINVAL);
 
 	if (!pskb_may_pull(skb, sizeof(*esph) + crypto_aead_ivsize(aead)))
-		goto out;
+		return ERR_PTR(-EINVAL);
 
 	__skb_pull(skb, sizeof(*esph) + crypto_aead_ivsize(aead));
 
 	skb->encap_hdr_csum = 1;
 
-	if (!(features & NETIF_F_HW_ESP))
+	if (!(features & NETIF_F_HW_ESP) || !x->xso.offload_handle ||
+	    (x->xso.dev != skb->dev))
 		esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
 
-	segs = x->outer_mode->gso_segment(x, skb, esp_features);
-	if (IS_ERR_OR_NULL(segs))
-		goto out;
-
-	__skb_pull(skb, skb->data - skb_mac_header(skb));
-
-	skb2 = segs;
-	do {
-		struct sk_buff *nskb = skb2->next;
-
-		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;
-
-		x->outer_mode->xmit(x, skb2);
-
-		err = x->type_offload->xmit(x, skb2, esp_features);
-		if (err) {
-			kfree_skb_list(segs);
-			return ERR_PTR(err);
-		}
-
-		if (!skb_is_gso(skb2))
-			seq++;
-		else
-			seq += skb_shinfo(skb2)->gso_segs;
-
-		skb_push(skb2, skb2->mac_len);
-		skb2 = nskb;
-	} while (skb2);
+	xo->flags |= XFRM_GSO_SEGMENT;
 
-out:
-	return segs;
+	return x->outer_mode->gso_segment(x, skb, esp_features);
 }
 
 static int esp6_input_tail(struct xfrm_state *x, struct sk_buff *skb)
@@ -222,6 +183,7 @@ static int esp6_input_tail(struct xfrm_state *x, struct sk_buff *skb)
 
 static int esp6_xmit(struct xfrm_state *x, struct sk_buff *skb,  netdev_features_t features)
 {
+	int len;
 	int err;
 	int alen;
 	int blksize;
@@ -230,6 +192,7 @@ static int esp6_xmit(struct xfrm_state *x, struct sk_buff *skb,  netdev_features
 	struct crypto_aead *aead;
 	struct esp_info esp;
 	bool hw_offload = true;
+	__u32 seq;
 
 	esp.inplace = true;
 
@@ -265,28 +228,33 @@ static int esp6_xmit(struct xfrm_state *x, struct sk_buff *skb,  netdev_features
 			return esp.nfrags;
 	}
 
+	seq = xo->seq.low;
+
 	esph = ip_esp_hdr(skb);
 	esph->spi = x->id.spi;
 
 	skb_push(skb, -skb_network_offset(skb));
 
 	if (xo->flags & XFRM_GSO_SEGMENT) {
-		esph->seq_no = htonl(xo->seq.low);
-	} else {
-		int len;
-
-		len = skb->len - sizeof(struct ipv6hdr);
-		if (len > IPV6_MAXPLEN)
-			len = 0;
+		esph->seq_no = htonl(seq);
 
-		ipv6_hdr(skb)->payload_len = htons(len);
+		if (!skb_is_gso(skb))
+			xo->seq.low++;
+		else
+			xo->seq.low += skb_shinfo(skb)->gso_segs;
 	}
 
+	esp.seqno = cpu_to_be64(xo->seq.low + ((u64)xo->seq.hi << 32));
+
+	len = skb->len - sizeof(struct ipv6hdr);
+	if (len > IPV6_MAXPLEN)
+		len = 0;
+
+	ipv6_hdr(skb)->payload_len = htons(len);
+
 	if (hw_offload)
 		return 0;
 
-	esp.seqno = cpu_to_be64(xo->seq.low + ((u64)xo->seq.hi << 32));
-
 	err = esp6_output_tail(x, skb, &esp);
 	if (err)
 		return err;
diff --git a/net/ipv6/xfrm6_mode_tunnel.c b/net/ipv6/xfrm6_mode_tunnel.c
index e66b94f46532..4e12859bc2ee 100644
--- a/net/ipv6/xfrm6_mode_tunnel.c
+++ b/net/ipv6/xfrm6_mode_tunnel.c
@@ -105,17 +105,14 @@ static struct sk_buff *xfrm6_mode_tunnel_gso_segment(struct xfrm_state *x,
 {
 	__skb_push(skb, skb->mac_len);
 	return skb_mac_gso_segment(skb, features);
-
 }
 
 static void xfrm6_mode_tunnel_xmit(struct xfrm_state *x, struct sk_buff *skb)
 {
 	struct xfrm_offload *xo = xfrm_offload(skb);
 
-	if (xo->flags & XFRM_GSO_SEGMENT) {
-		skb->network_header = skb->network_header - x->props.header_len;
+	if (xo->flags & XFRM_GSO_SEGMENT)
 		skb->transport_header = skb->network_header + sizeof(struct ipv6hdr);
-	}
 
 	skb_reset_mac_len(skb);
 	pskb_pull(skb, skb->mac_len + x->props.header_len);
diff --git a/net/xfrm/xfrm_device.c b/net/xfrm/xfrm_device.c
index c61a7d46b412..d9fe06da21ac 100644
--- a/net/xfrm/xfrm_device.c
+++ b/net/xfrm/xfrm_device.c
@@ -23,32 +23,99 @@
 #include <linux/notifier.h>
 
 #ifdef CONFIG_XFRM_OFFLOAD
-int 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 err;
+	__u32 seq;
 	struct xfrm_state *x;
+	struct sk_buff *skb2;
+	netdev_features_t esp_features = features;
 	struct xfrm_offload *xo = xfrm_offload(skb);
 
-	if (skb_is_gso(skb))
-		return 0;
+	if (!xo)
+		return skb;
 
-	if (xo) {
-		x = skb->sp->xvec[skb->sp->len - 1];
-		if (xo->flags & XFRM_GRO || x->xso.flags & XFRM_OFFLOAD_INBOUND)
-			return 0;
+	if (!(features & NETIF_F_HW_ESP))
+		esp_features = features & ~(NETIF_F_SG | NETIF_F_CSUM_MASK);
+
+	x = skb->sp->xvec[skb->sp->len - 1];
+	if (xo->flags & XFRM_GRO || x->xso.flags & XFRM_OFFLOAD_INBOUND)
+		return skb;
+
+	if (skb_is_gso(skb)) {
+		struct net_device *dev = skb->dev;
+
+		if (unlikely(!x->xso.offload_handle || (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);
 
+			segs = skb_gso_segment(skb, esp_features);
+			if (IS_ERR(segs)) {
+				XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
+				kfree_skb(skb);
+				return NULL;
+			} else {
+				consume_skb(skb);
+				skb = segs;
+			}
+		} else {
+			return skb;
+		}
+	}
+
+	if (!skb->next) {
 		x->outer_mode->xmit(x, skb);
 
-		err = x->type_offload->xmit(x, skb, features);
+		err = x->type_offload->xmit(x, skb, esp_features);
 		if (err) {
 			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
-			return err;
+			kfree_skb(skb);
+			return NULL;
 		}
 
 		skb_push(skb, skb->data - skb_mac_header(skb));
+
+		return skb;
 	}
 
-	return 0;
+	skb2 = skb;
+	seq = xo->seq.low;
+
+	do {
+		struct sk_buff *nskb = skb2->next;
+
+		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;
+
+		x->outer_mode->xmit(x, skb2);
+
+		err = x->type_offload->xmit(x, skb2, esp_features);
+		if (err) {
+			XFRM_INC_STATS(xs_net(x), LINUX_MIB_XFRMOUTSTATEPROTOERROR);
+			skb2->next = nskb;
+			kfree_skb_list(skb2);
+			return NULL;
+		}
+
+		if (!skb_is_gso(skb2))
+			seq++;
+		else
+			seq += skb_shinfo(skb2)->gso_segs;
+
+		skb_push(skb2, skb2->data - skb_mac_header(skb2));
+
+		skb2 = nskb;
+	} while (skb2);
+
+	return skb;
 }
 EXPORT_SYMBOL_GPL(validate_xmit_xfrm);
 
-- 
2.14.1

[PATCH RFC v2 2/5] net: Add asynchronous callbacks for xfrm on layer 2.

[Steffen Klassert]

This patch implements asynchronous crypto callbacks
and a backlog handler that can be used when IPsec
is done at layer 2 in the TX path. It also extends
the skb validate functions so that we can update
the driver transmit return codes based on async
crypto operation or to indicate that we queued the
packet in a backlog queue.

Joint work with: Aviv Heller <avivh@mellanox.com>

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/linux/netdevice.h |   6 ++-
 include/net/xfrm.h        |  22 ++++++++--
 net/core/dev.c            |  16 +++++---
 net/ipv4/esp4.c           |  24 +++++++++--
 net/ipv6/esp6.c           |  24 +++++++++--
 net/packet/af_packet.c    |   3 +-
 net/sched/sch_generic.c   |  16 +++++++-
 net/xfrm/xfrm_device.c    | 100 +++++++++++++++++++++++++++++++++++++---------
 8 files changed, 175 insertions(+), 36 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index cc4ce7456e38..c82d207ebc97 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -2793,7 +2793,9 @@ 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.
@@ -3325,7 +3327,7 @@ 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, bool *again);
 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 df7f3d0ac4a1..2517c4f7781a 100644
--- a/include/net/xfrm.h
+++ b/include/net/xfrm.h
@@ -1051,6 +1051,7 @@ 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
@@ -1874,21 +1875,28 @@ static inline struct xfrm_state *xfrm_input_state(struct sk_buff *skb)
 {
 	return skb->sp->xvec[skb->sp->len - 1];
 }
+#endif
+
 static inline struct xfrm_offload *xfrm_offload(struct sk_buff *skb)
 {
+#ifdef CONFIG_XFRM
 	struct sec_path *sp = skb->sp;
 
 	if (!sp || !sp->olen || sp->len != sp->olen)
 		return NULL;
 
 	return &sp->ovec[sp->olen - 1];
-}
+#else
+	return NULL;
 #endif
+}
 
 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, bool *again);
 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);
@@ -1929,7 +1937,15 @@ 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, bool *again)
 {
 	return skb;
 }
diff --git a/net/core/dev.c b/net/core/dev.c
index 7b89cbd9ef4c..ca2c1b93ec45 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3042,7 +3042,7 @@ 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, bool *again)
 {
 	netdev_features_t features;
 
@@ -3082,7 +3082,7 @@ 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, again);
 
 	return skb;
 
@@ -3093,7 +3093,7 @@ 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, bool *again)
 {
 	struct sk_buff *next, *head = NULL, *tail;
 
@@ -3104,7 +3104,7 @@ 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, again);
 		if (!skb)
 			continue;
 
@@ -3431,6 +3431,7 @@ static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
 	struct netdev_queue *txq;
 	struct Qdisc *q;
 	int rc = -ENOMEM;
+	bool again = false;
 
 	skb_reset_mac_header(skb);
 
@@ -3492,7 +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, &again);
 			if (!skb)
 				goto out;
 
@@ -4176,6 +4177,8 @@ 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)
@@ -8844,6 +8847,9 @@ 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 d57aa64fa7c7..7948833dc204 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 a902ff8f59be..08a424fa8009 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 da215e5c1399..ee7aa0ba3a67 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -247,12 +247,13 @@ static int packet_direct_xmit(struct sk_buff *skb)
 	struct sk_buff *orig_skb = skb;
 	struct netdev_queue *txq;
 	int ret = NETDEV_TX_BUSY;
+	bool again = false;
 
 	if (unlikely(!netif_running(dev) ||
 		     !netif_carrier_ok(dev)))
 		goto drop;
 
-	skb = validate_xmit_skb_list(skb, dev);
+	skb = validate_xmit_skb_list(skb, dev, &again);
 	if (skb != orig_skb)
 		goto drop;
 
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 981c08fe810b..f51e481a54c2 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -32,6 +32,7 @@
 #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;
@@ -230,6 +231,8 @@ static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
 
 		/* 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)) {
@@ -285,6 +288,7 @@ bool sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
 		     spinlock_t *root_lock, bool validate)
 {
 	int ret = NETDEV_TX_BUSY;
+	bool again = false;
 
 	/* And release qdisc */
 	if (root_lock)
@@ -292,7 +296,17 @@ bool 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, &again);
+
+#ifdef CONFIG_XFRM_OFFLOAD
+	if (unlikely(again)) {
+		if (root_lock)
+			spin_lock(root_lock);
+
+		dev_requeue_skb(skb, q);
+		return false;
+	}
+#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 d9fe06da21ac..3f91ccffbcc8 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, bool *again)
 {
 	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 +43,16 @@ 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;
 
+	local_irq_save(flags);
+	sd = this_cpu_ptr(&softnet_data);
+	err = !skb_queue_empty(&sd->xfrm_backlog);
+	local_irq_restore(flags);
+
+	if (err) {
+		*again = true;
+		return skb;
+	}
+
 	if (skb_is_gso(skb)) {
 		struct net_device *dev = skb->dev;
 
@@ -54,23 +65,26 @@ 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 +96,37 @@ 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)
+				return NULL;
 
-		if (!skb_is_gso(skb2))
-			seq++;
-		else
-			seq += skb_shinfo(skb2)->gso_segs;
+			goto skip_push;
+		}
 
 		skb_push(skb2, skb2->data - skb_mac_header(skb2));
 
+skip_push:
 		skb2 = nskb;
 	} while (skb2);
 
@@ -207,6 +222,55 @@ 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)
+{
+	struct net_device *dev = skb->dev;
+	int ret = NETDEV_TX_BUSY;
+	struct netdev_queue *txq;
+	struct softnet_data *sd;
+	unsigned long flags;
+
+	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_head *xfrm_backlog = &sd->xfrm_backlog;
+	struct sk_buff_head list;
+	struct sk_buff *skb;
+
+	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)
-- 
2.14.1

[PATCH RFC v2 3/5] xfrm: Allow to use the layer2 IPsec GSO codepath for software crypto.

[Steffen Klassert]

We now have support for asynchronous crypto operations in the layer 2 TX
path. This was the missing part to allow the GSO codepath for software
crypto, so allow this codepath now.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 net/xfrm/xfrm_device.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/net/xfrm/xfrm_device.c b/net/xfrm/xfrm_device.c
index 3f91ccffbcc8..03f2bc61f1fc 100644
--- a/net/xfrm/xfrm_device.c
+++ b/net/xfrm/xfrm_device.c
@@ -202,8 +202,8 @@ bool xfrm_dev_offload_ok(struct sk_buff *skb, struct xfrm_state *x)
 	if (!x->type_offload || x->encap)
 		return false;
 
-	if ((x->xso.offload_handle && (dev == xfrm_dst_path(dst)->dev)) &&
-	     !xdst->child->xfrm && x->type->get_mtu) {
+	if ((!dev || (x->xso.offload_handle && (dev == xfrm_dst_path(dst)->dev))) &&
+	     (!xdst->child->xfrm && x->type->get_mtu)) {
 		mtu = x->type->get_mtu(x, xdst->child_mtu_cached);
 
 		if (skb->len <= mtu)
-- 
2.14.1

[PATCH RFC v2 4/5] xfrm: Allow IPsec GSO with software crypto for local sockets.

[Steffen Klassert]

With support of async crypto operations in the GSO codepath
we have everything in place to allow GSO for local sockets.
This patch enables the GSO codepath.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/net/xfrm.h | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index 2517c4f7781a..357764a2bb4e 100644
--- a/include/net/xfrm.h
+++ b/include/net/xfrm.h
@@ -1910,6 +1910,8 @@ static inline bool xfrm_dst_offload_ok(struct dst_entry *dst)
 		return false;
 
 	xdst = (struct xfrm_dst *) dst;
+	if (!x->xso.offload_handle && !xdst->child->xfrm)
+		return true;
 	if (x->xso.offload_handle && (x->xso.dev == xfrm_dst_path(dst)->dev) &&
 	    !xdst->child->xfrm)
 		return true;
-- 
2.14.1

[PATCH RFC v2 5/5] esp: Don't require synchronous crypto fallback on offloading anymore.

[Steffen Klassert]

We support asynchronous crypto on layer 2 ESP now.
So no need to force synchronous crypto fallback on
offloading anymore.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 net/ipv4/esp4.c | 12 ++----------
 net/ipv6/esp6.c | 12 ++----------
 2 files changed, 4 insertions(+), 20 deletions(-)

diff --git a/net/ipv4/esp4.c b/net/ipv4/esp4.c
index 7948833dc204..6f00e43120a8 100644
--- a/net/ipv4/esp4.c
+++ b/net/ipv4/esp4.c
@@ -843,17 +843,13 @@ static int esp_init_aead(struct xfrm_state *x)
 	char aead_name[CRYPTO_MAX_ALG_NAME];
 	struct crypto_aead *aead;
 	int err;
-	u32 mask = 0;
 
 	err = -ENAMETOOLONG;
 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
 		goto error;
 
-	if (x->xso.offload_handle)
-		mask |= CRYPTO_ALG_ASYNC;
-
-	aead = crypto_alloc_aead(aead_name, 0, mask);
+	aead = crypto_alloc_aead(aead_name, 0, 0);
 	err = PTR_ERR(aead);
 	if (IS_ERR(aead))
 		goto error;
@@ -883,7 +879,6 @@ static int esp_init_authenc(struct xfrm_state *x)
 	char authenc_name[CRYPTO_MAX_ALG_NAME];
 	unsigned int keylen;
 	int err;
-	u32 mask = 0;
 
 	err = -EINVAL;
 	if (!x->ealg)
@@ -909,10 +904,7 @@ static int esp_init_authenc(struct xfrm_state *x)
 			goto error;
 	}
 
-	if (x->xso.offload_handle)
-		mask |= CRYPTO_ALG_ASYNC;
-
-	aead = crypto_alloc_aead(authenc_name, 0, mask);
+	aead = crypto_alloc_aead(authenc_name, 0, 0);
 	err = PTR_ERR(aead);
 	if (IS_ERR(aead))
 		goto error;
diff --git a/net/ipv6/esp6.c b/net/ipv6/esp6.c
index 08a424fa8009..7c888c6e53a9 100644
--- a/net/ipv6/esp6.c
+++ b/net/ipv6/esp6.c
@@ -752,17 +752,13 @@ static int esp_init_aead(struct xfrm_state *x)
 	char aead_name[CRYPTO_MAX_ALG_NAME];
 	struct crypto_aead *aead;
 	int err;
-	u32 mask = 0;
 
 	err = -ENAMETOOLONG;
 	if (snprintf(aead_name, CRYPTO_MAX_ALG_NAME, "%s(%s)",
 		     x->geniv, x->aead->alg_name) >= CRYPTO_MAX_ALG_NAME)
 		goto error;
 
-	if (x->xso.offload_handle)
-		mask |= CRYPTO_ALG_ASYNC;
-
-	aead = crypto_alloc_aead(aead_name, 0, mask);
+	aead = crypto_alloc_aead(aead_name, 0, 0);
 	err = PTR_ERR(aead);
 	if (IS_ERR(aead))
 		goto error;
@@ -792,7 +788,6 @@ static int esp_init_authenc(struct xfrm_state *x)
 	char authenc_name[CRYPTO_MAX_ALG_NAME];
 	unsigned int keylen;
 	int err;
-	u32 mask = 0;
 
 	err = -EINVAL;
 	if (!x->ealg)
@@ -818,10 +813,7 @@ static int esp_init_authenc(struct xfrm_state *x)
 			goto error;
 	}
 
-	if (x->xso.offload_handle)
-		mask |= CRYPTO_ALG_ASYNC;
-
-	aead = crypto_alloc_aead(authenc_name, 0, mask);
+	aead = crypto_alloc_aead(authenc_name, 0, 0);
 	err = PTR_ERR(aead);
 	if (IS_ERR(aead))
 		goto error;
-- 
2.14.1

Re: [PATCH RFC v2 0/5] Support asynchronous crypto for IPsec GSO.

[Steffen Klassert]

On Fri, Dec 15, 2017 at 10:13:03AM +0100, Steffen Klassert wrote:
> This patchset implements asynchronous crypto handling
> in the layer 2 TX path. With this we can allow IPsec
> ESP GSO for software crypto. This also merges the IPsec
> GSO and non-GSO paths to both use validate_xmit_xfrm().
> 
> 1) Separate ESP handling from segmentation for GRO packets.
>    This unifies the IPsec GSO and non GSO codepath.
> 
> 2) Add asynchronous callbacks for xfrm on layer 2. This
>    adds the necessary infrastructure to core networking.
> 
> 3) Allow to use the layer2 IPsec GSO codepath for software
>    crypto, all infrastructure is there now.
> 
> 4) Also allow IPsec GSO with software crypto for local sockets.
> 
> 5) Don't require synchronous crypto fallback on IPsec offloading,
>    it is not needed anymore.
> 
> I plan to merge this to ipsec-next during this development cycle.

This is now applied to ipsec-next.